comprehensive evaluation of polar weather research and...

Comprehensive evaluation of polar weather researchand forecasting model performance in the Antarctic

David H Bromwich12 Francis O Otieno1 Keith M Hines1 Kevin W Manning3

and Elad Shilo4

Received 22 May 2012 revised 7 November 2012 accepted 8 November 2012 published 17 January 2013

[1] Recent versions of the Polar Weather Research and Forecasting model are evaluatedover the Antarctic to assess the impact of model improvements resolution large-scalecirculation variability and uncertainty in initial and lateral boundary conditions The modelskill differs more between forecasts using different sources of lateral boundary data thanbetween forecasts from different model versions or simulated years Using the ERA-Interimreanalysis for initial and lateral boundary conditions produces the best skill The forecastshave a cold summer and a warm winter bias in 2m air temperatures with similar butsmaller bias in dew point temperatures Upper air temperature biases are small and remainless than 1 C except at the tropopause in summer Geopotential height biases increasewith height in both seasons Deficient downward longwave radiation in all seasons and anunder representation of clouds enhance radiative loss leading to the cold summer biasExcess summer surface incident shortwave radiation plays a secondary role because 80 ofit is reflected leading to greater skill for clear compared with cloudy skies The positivewind speed bias produces a warm surface bias in winter resulting from anomalously largedownward flux of sensible heat toward the surface Low temperatures on the continent limitsublimation and hence the precipitable water amounts over the ice sheet ERA-Interimexperiments with higher precipitable water showed reduced biases in downwellingshortwave and longwave radiation Increasing horizontal resolution from 60 to 15 kmimproves the skill of surface wind forecasts

Citation Bromwich D H F O Otieno K M Hines K W Manning and E Shilo (2013) Comprehensive evaluationof polar weather research and forecasting model performance in the Antarctic J Geophys Res Atmos 118 274ndash292doi1010292012JD018139

1 Introduction

[2] As a result of advances in modeling observational datacoverage and communication systems critical forecasts insupport of the US Antarctic Program (USAP) are now rou-tinely made using the polar-optimized version of the WeatherResearch and Forecasting model (PolarWRF) via the AntarcticMesoscale Prediction System (AMPS) [Powers et al 2012]AMPS forecasts with Polar WRF version 301 began inAugust 2008 and the model was used with few configurationchanges until April 2011 when an upgrade to Polar WRF321 was implemented Not only did the AMPS model change

but as computer capability has increased also has the resolu-tion used by AMPS which now stands at 167 km aroundMcMurdo Station where the USAP hub is located Howeverthe WRF model was originally developed for the NorthernHemisphere midlatitudes and for polar regions WRF has beentested primarily in Arctic environments [Bromwich et al2009Hines and Bromwich 2008Hines et al 2011Cassanoet al 2011] Therefore it is necessary for advancing Antarcticmodeling to benchmark its performance in Antarctica[3] Relying on these AMPS forecasts six ski-equipped

LC-130 Hercules aircraft flown by the New York AirNational Guard completed 359 missions while the 62nd and446th Airlift Wing flew 72 missions in support of OperationDeep Freeze during the 2011ndash2012 field season togetherthey transported at least 6600 people and 58 million kilo-grams of equipment [Hannon 2012] In addition to transportlogistics accurate forecasts are also needed for safety on theground A number of people have had to be evacuated atvarious times from the continent as recently as August2012 [National Science Foundation (NSF) 2012] and tour-ist visits now exceed 30000 annually [International Associ-ation of Antarctic Tour Operators (IAATO) 2010] Accurateforecasts will therefore continue to be required for the safetyof those working in Antarctica during severe wind storms

1Polar Meteorology Group Byrd Polar Research Center The Ohio StateUniversity Columbus Ohio USA

2Atmospheric Sciences Program Department of Geography The OhioState University Columbus Ohio USA

3National Center for Atmospheric Research Boulder Colorado USA4Israeli Meteorological Service Bet Dagan Israel

Corresponding author D H Bromwich Polar Meteorology GroupByrd Polar Research Center The Ohio State University 1090 CarmackRoad Columbus OH 43210 USA (bromwich1osuedu45)

copy2012 American Geophysical Union All Rights Reserved2169-897X132012JD018139

274

JOURNAL OF GEOPHYSICAL RESEARCH ATMOSPHERES VOL 118 274ndash292 doi1010292012JD018139 2013

and in potential large-scale evacuations in the future In addi-tion archived AMPS forecasts have been used extensively tostudy Antarctic weather and climate [eg Monaghan et al2003 2005 Fogt and Bromwich 2008 Powers 2007Powers et al 2010 2012 Steinhoff 2011 Bromwich et al2011a Nigro et al 2012 Seefeldt and Cassano 2012Steinhoff et al 2012][4] Model simulations produce data that in the absence of

long-term observations can be used to diagnose climaticprocesses such as those depicted in Antarctic ice corescovering eight glacial cycles [European Project for Ice Cor-ing in Antarctica (EPICA) 2004] Such applications willrequire climate models that accurately predict the short-termweather (forecasts) before they can be used to diagnoselong-term climate process such the accelerated rate of iceloss from the continent [Velicogna 2009 Rignot et al2011] Obtaining a complete understanding of Antarcticteleconnections and climate variability using models is anarea of active research and a recent report of the NationalResearch Council (NRC) [2011] supports the developmentof a new generation of robust earth system models[5] The skill of Polar WRF already demonstrated in the

Arctic may not prevail in Antarctica because of substantialdifferences between the two polar regions First the atmo-spheric circulation in Antarctica is strongly influenced bythe presence of the large ice sheet with elevations exceeding3 km in most areas Any mesoscale model used here mustaccurately represent the strong katabatic winds that aregoverned by the balance of gravity thermal stability andsynoptic forcing [eg Bromwich and Liu 1996 Cassanoand Parish 2000 Parish and Bromwich 2007] The coldtemperatures limit evaporation and sublimation and makeice the dominant phase of water in Antarctic clouds Highsurface albedo [Laine 2008] and relatively pristine skies(low amounts of aerosol) also result in different radiation-atmosphere interactions Surrounded by a large oceanAntarctica has limited observational data needed to initializemesoscale models and to validate forecasts making the veri-fication of models more difficult than in the Arctic[6] Development of the standard WRF is a community

effort led by the National Center for Atmospheric Research(NCAR) among others Benefits of community involvementin the development include rapid and continuous refinementsto both the standard and the Polar WRF The primary draw-back to such rapid developments is that newer model ver-sions are subject to fewer evaluations than their predecessorsAll past verifications cannot be repeated every time a newmodel is released Incompatibilities in the physics parameter-izations and other errors (bugs) also arise occasionally fromthe model developments themselves In a recent Arctic-wideevaluation Polar WRF 311 demonstrated significant Arcticforecast skill This study expands the evaluations to all ofAntarctica on an annual time scale and is the sixth in a series[Hines and Bromwich 2008 Bromwich et al 2009 Hineset al 2011Wilson et al 2011 2012] of verification studiesdocumenting ongoing Polar WRF developments We focuson three factors that influence forecast skill model improve-ments uncertainty in driving data and interannual varia-tions in the large-scale atmospheric circulation The impactof model improvements is assessed using four recent ver-sions of Polar WRF (releases 301 311 321 and 331)Bug-fixed subversions are chosen because they include

corrections for problems identified by the user communitysince the initial release Improvements that only make themodel easy to use enhance computational efficiency orspeed up diagnosis of model output are not considered[7] This study examines the impact on forecast skill of un-

certainty in the initial and lateral boundary conditions bycomparing Polar WRF forecasts made using the reanalysisfrom the European Centre for Medium-Range Weather Fore-casts (ECMWF ERA-Interim) [Dee et al 2011] and thefinal analysis from the National Centers for EnvironmentalPrediction (NCEP) operational global analysis (GFS-FNL)The relative accuracy of contemporary reanalyses has beenexamined previously in relation to predicted precipitationand ERA-Interim fared the best [Nicolas and Bromwich2011 Bromwich et al 2011b] In addition from a compar-ison of five reanalyses with Antarctic station observations[Bracegirdle and Marshall 2012] ERA-Interim was foundto be the most reliable for mean sea level pressure and500 hPa height trends ERA-Interim represents a majoradvance over previous ECMWF reanalyses because ofincreased resolution (T255) a more advanced atmosphericmodel 4D-Var assimilation system and the enormous vol-ume of assimilated data Assimilation of satellite radiancedata is especially critical over the Southern Ocean wherethere are few synoptic stations However reanalyses suchas ERA-Interim have the advantage of a fixed modeling sys-tem and input data that are usually unavailable in operationalforecasting Thus many centers rely on operational analysesfor lateral boundary conditions For this reason this studycompares Polar WRF skill when driven by an analysis anda reanalysis The two cover a reasonable range that can beassociated with the uncertainty in the lateral boundary dataand are considered sufficient for the objectives here How-ever we note that many other sources of lateral and initialconditions exist[8] Influence of interannual variations in the large-scale

atmospheric circulation on model skill is evaluated by com-paring forecast statistics from two different years (1993 and2007) both made using the same analysis (ERA-Interim)We choose 1993 to allow for a comparison between thestatistics given here with those of an earlier Polar MM5 Ant-arctic verification [Guo et al 2003] and using 2007 allowsus to take advantage of recent advances in the observingnetwork Strictly speaking the quality of a reanalysisdepends on the number of assimilated observations andthese did differ between 1993 and 2007 thereby compound-ing differences resulting from interannual variations withthose resulting from analysis quality However weassume that ERA-Interim representations of Antarctic atmo-spheric circulation are among the best possible for these years[Bromwich et al 2011b]

2 Model Experiments Lower BoundaryConditions and Observations

21 Model Description

[9] Polar WRF [Hines and Bromwich 2008 Hines et al2011] was developed by the Polar Meteorology Group atThe Ohio State University and builds on previous successwith an earlier mesoscale model (Polar MM5) [Bromwichet al 2001] in the Arctic The overall difference is that PolarWRF is modified with enhancements that adapt the standard

BROMWICH ET AL POLAR WEATHER RESEARCH AND FORECASTING MODEL

275

code to the polar regions These include freezing point ofseawater set at 27136K surface roughness of 0001m oversea ice and permanent land ice snow emissivity of 098snow density over sea ice of 300 kg3 and use of a verticaldensity profile over land ice The thermal conductivity of thetransition layer between the atmosphere and snow is set tothat of snow and whenever the upper snow layer exceeds20 cm depth it is treated as if the snow was 20 cm thickfor heat calculations In addition sea ice albedo sea icethickness and snow cover on sea ice are fixed in the stan-dard WRF but can vary in Polar WRF since version 321For consistency between Polar WRF versions these lastoptions are not used here The distinction between the twocodes has recently been blurred somewhat as some PolarWRF modifications in particular the treatment of sea icethrough the mosaic method [Bromwich et al 2009] andchanges to the thermal properties of snow and ice [Hinesand Bromwich 2008] have been incorporated into the stan-dard release from NCAR[10] Both models use fully compressible Euler nonhydro-

static equations for atmospheric dynamics on a horizontalArakawa C-grid staggering and a vertical terrain-followinghydrostatic-pressure coordinate Because it is a limited-areamodel Polar WRF requires information on the wind temper-ature geopotential and relative humidity at the lateralboundaries At the lower boundary the surface elevationpressure sea ice conditions and sea surface temperaturesalso have to be specified Reflection of gravity waves fromthe upper boundary is problematic and is controlled by eithera layer of increased diffusion Rayleigh relaxation or theinclusion of a layer with implicit gravity-wave damping[Chini and Leibovich 2003] In this study increased diffu-sion and damping of the vertical velocity over an 8 km depth(approximately four levels) from the model top are appliedA more comprehensive description of WRF is availablefrom Skamarock et al [2009] and the WRF userrsquos guide

(httpwwwmmmucareduwrfusersdocsuser_guide_V3ARWUsers-GuideV3pdf)[11] Improvements in both models can be grouped under

physics dynamics nudging initialization and software Witheach version ofWRF not only the details of the improvementsare provided but also a list of known problems Only a sum-mary of improvements relevant to this study is presented hereFor a complete list of updates from WRF version 301 to331 the reader should go the WRF userrsquos website (httpwwwmmmucareduwrfusers) Table 1 shows that the mainchange in the model versions considered here are changes tothe radiation scheme however there are other differences inthe model versions that are not immediately evident InWRF 301 for example an option was added that allowssea ice fraction and albedo to be updated and some changeswere made to emissivity values In version 311 the Mellor-Yamada-Janjic (MYJ) planetary boundary layer (PBL) andMYJ surface physics were updated to the NCEP operationalversions and some minor fixes were made for polar fastFourier transforms The Mellor-Yamada Nakanishi Niino(MYNN) level 25 and level 3 PBL and MYNN surfaceschemes were also added Therefore in WRF 311 one hasthe option of using MYJ or MYNN for the PBL and surfacephysics A fix for a cooling problem near model top thatresulted from using vertically interpolated relative humidityrather than mixing ratio was made in WRF 32 Although allthe experiments described here use the Grell-Devenyi cumulusscheme a minor fix was added for vertical shear calculation inversion 33

22 Experimental Strategy

[12] In assessing the impact of model improvements neweroptions for the same physics schemes are usedwhenever avail-able (see Table 1 for a summary of the physics packages used)The RRTM longwave scheme [Mlawer et al 1997] is used inPolar WRF 301 and 311 whereas the RRTM for generalcirculation models (RRTMG) is used in Polar WRF 321and 331 Use of the RRTMG shortwave radiation scheme isa requirement for using the new RRTMG longwave schemeRRTMG accounts for all major gaseous absorbers at longand short wavelengths along with the radiative effects ofclouds and aerosols as well as a statistical technique for repre-senting subgrid-scale cloud variabilityHines et al [2011] re-cently performed four simulations to assess the sensitivity ofthe various microphysics options in Polar WRF and foundonly a small sensitivity over Arctic land Thus in Antarctica(which is mostly land ice) the impact of using the WRF Sin-gle Moment 5-class microphysics (WSM5) instead of theMorrison scheme that is used in the Arctic System Reanalysisis presumed to be small AMPS also uses the WSM5 so thelevel of skill found here provides a benchmark for the compa-rable resolution of AMPS experimental forecasts[13] There are three full-year experiments using GFS-FNL

analysis for 2007 but each was conducted with a differentversion of Polar WRF (obtained from NCAR [1999]) In addi-tion a total of seven sensitivity experiments described furtherin section 32 are performed for January (summer) and July(winter) 2007 They are used to assess sensitivity to lateralboundary conditions (GFS-FNL versus ERA-Interim) radia-tion physics and planetary boundary layer schemes Thechoice of physics experiments was based on known differ-ences in the physical characteristics between Antarctica and

Table 1 The WRF Configuration and Physics Options Used

Model VersionPhysics 301 311 321

LongwaveRRTM X XRRTMG X

ShortwaveGoddard X XRRTMG X

Microphysics WRF Single moment 5 classLateral boundary NCEP Global Final analysis

GFS-FNL 2007)Planetary boundary layer Mellor Yamada-Janjic (MYJ)Sea surface temperature (SST) NCEP RTG (2007)Land surface Noah Land Surface Model (LSM)Surface layer Monin-Obukhov (Janjic-eta)Time step 120 sHorizontal resolution 60 kmCumulus parameterization Grell-DevenyiModel top 10 hPa with damping in the top four

levels (8 km depth)LatitudeLongitude 121 121Relaxation zone 10 grid pointsVertical resolution 39 eta levelsSea ice NSIDC 25 km fractional sea iceIntegration 48 h forecasts from global analysisSpin up First 24 h used for model spin upBase state temperature 27316K


276

the Arctic Antarctic radiation balance is likely differentbecause the clouds are composed primarily of ice and becauseof the high surface albedo In addition the thick ice sheetestablishes an extremely stable boundary layer profile whichinfluences near-surface fluxes Therefore sensitivity to thePBL and radiation parameterizations were identified as beingvery good candidates for this study[14] The analysis and reanalysis described earlier are used

to specify lateral boundary conditions on the five outermost

grid points and gradually relaxed over the next 10 resultingin a 15-point forcing frame For each day of the year themodel is initialized with a 24 h lead time using analysis orreanalysis values (depending on the experiment) and inte-grated continuously for 48 h Only forecasts from the last24 h from each of the forecasts for the consecutive calendardays are joined to make an annual time series from whichthe monthly analyses are performed in section 30 Usingonly the last 24 h from each forecast allows for spin-up ofthe hydrological cycle and planetary boundary layer structureover the first 24 h of the 48 h forecast Guo et al [2003] useda 72h integration period but our preliminary analysis showedthat the centers of forecast Antarctic storms drift significantlywithout nudging in longer-duration simulations hence theshorter duration is used here[15] Because of the amount of simulations required a sin-

gle domain comparable to that used by Guo et al [2003] andcentered at the South Pole with the top at 10 hPa 39 verticallevels and a 60 km horizontal resolution is used (Figure 1a)However at this resolution a number of Antarctic stationslocated on the steep coastal slopes and some isolated islandsare not properly represented For example Base Orcadas(6074S 4474W) is a land point 8m above mean sealevel yet at this resolution the nearest grid point is locatedover water at sea level The impact of the coarse resolutionon the statistics is minimized in this study by using onlystations whose surface elevation in the model does not differfrom the actual station height by more than 04 km (Figure 1b)and whose weather is not strongly influenced by local featuressuch as topography For analysis the forecasts are interpolatedto the station sites using a bilinear interpolation while avoid-ing mixing land and ocean grid points because of their differ-ent diurnal characteristics A higher resolution is achievablethrough nesting and could downscale the results to less than2 km (as in AMPS Grid 5) but it is computationally expen-sive for our domain size integration duration and numberof experiments Furthermore high forecast skill on the parentdomain is critical for accuracy on any of the higher resolutionnests Nevertheless 1month sensitivity experiments duringboth summer and winter are used here to assess the impactof a fourfold increase in model resolution on forecast skill(section 32)

23 Lower Boundary Conditions

[16] Lower boundary conditions are specified by terrainelevation sea ice concentrations and sea surface tempera-tures (SST) Elevation data from the 200m Radarsat Antarc-tic Mapping Project Digital Elevation Model (RAMP-DEM)[Liu et al 2001] are interpolated using standard WRF rou-tines to the domain in Figure 1a Fractional sea ice concen-trations are specified from passive microwave measurementsfrom DMSP Special Sensor Microwave Imagers (SMMI)available from the National Snow Ice Data Center (NSIDC)[Comiso 2007] and are based on the bootstrap algorithmAlthough the sources of terrain and sea ice concentration re-main the same for all experiments the SST data are eitherspecified from the archives of NCEP 05 real-time global(RTG) SST [Gemmill et al 2007] analyses for 2007 or theNOAA weekly 10 optimum interpolated observations(OI) [Reynolds et al 2007] for 1993 (prior to RTG SSTs)Both SSTs and sea ice fraction concentrations are linearly

Figure 1 (a) Location of the observational sites superim-posed on the model topography (250 m interval) andobserved average sea ice extent (blue contour) depicted usingthe 001 fractional sea ice concentration contour The blueblack orange and red dots denote Automatic Weather(AWS) synoptic (NCDC) radiation (BSRN) and IGRA up-per air stations respectively Ice shelves are shaded gray (b)Station elevation (y axis) and elevation errors (x axis) at60 km (black dots) and at 15 km (red asterisks) model resolu-tion some of the stations below 05 km are displaced verti-cally in the model by nearly 1 km Stations not properlyrepresented in the model (height errors exceeding 400m in-dicated by vertical lines) at 60 km have been excluded fromthe statistical analysis


277

interpolated in time to update the lower boundary conditionsevery 6 h throughout the 48 h of forecast[17] Unlike a case in the midlatitudes where many soil

temperature measurements exist Antarctica has few ob-served snowpack profiles needed to specify accurately thesnowpack temperature profiles Weller and Schwerdtfeger[1977] measured the thermal properties of the uppermost10m of snow at a site on the Antarctic Plateau The resultsshow that the monthly temperature near the surface has astrong seasonal cycle whereas at the 10m depth the temper-ature is equal to the annual mean Therefore this studyassumes that the stable stratification over the permanent icecover produces ice surface temperatures that are roughly inequilibrium with the air at 2m above the icesnow surfaceMonthly mean 2m air temperatures are therefore used tospecify the temperature at the top of the snowpack and cli-matological annual mean is used to specify temperatures atthe bottom of the snowpack as required at a depth of 8min the Noah LSM [Chen and Dudhia 2001] Temperaturesfor intermediate depths are linearly interpolated from thesetwo

24 Observational Data

[18] The sites of observations used for the model compari-son are given in Figure 1a Synoptic station observations areretrieved from the National Climate Data Center (NCDC)archives and augmented by Automatic Weather Station(AWS) observations from the University of Wisconsinrsquos Ant-arctic Meteorological Research Center (AMRC) the ItalianAntarctic Research Program and the Australian AntarcticData Center AWS units measure air temperature wind speedand wind direction at ~3m whereas the other variables aremeasured at ~1m Therefore for AWS sites the forecast10mwind speeds are adjusted using the logarithmic wind pro-file down to 3m above the model surface to correspond moreclosely to the AWS wind measurements Measurements of thesurface downwelling shortwave and longwave radiation fromthe Baseline Surface Radiation Network (BSRN) [Ohmuraet al 1998] of Antarctic Stations [Dutton 2008 Vitale2009 Yamanouchi 2010 Koumlnig-Langlo 2011] are used inthe surface energy budget analysis Above the surface upperair measurements from the British Antarctic Survey and theIntegrated Global Radiosonde Archive (IGRA) [Durre et al2006 2008] are used to evaluate the forecasts[19] The extreme climate and challenging environment in

which Antarctic observations are made inevitably producesome unusable measurements Therefore very stringent qualitycontrol measures are necessary to isolate good observationsFirst a visual examination individual time series for suspi-cious characteristics is performed for each station and variableThese include for example large spikes in surface pressure(eg gt50 hPa over a 10min period) a zero wind speed read-ing with non-missing or fixed direction etc Because of diffi-culties in distinguishing AWS units reporting calm conditionsfrom those with frozen anemometers all wind speeds less than01m s1 are treated as missing in this study Next the avail-ability (gt50) and representativeness (lt3s) are examinedMonthly statistics are used in the quality control to accommo-date Antarctic summer-only stations with extensive summerobservations but with large gaps at other times Finally thenumber of stations is thinned out to prevent areas such as theRoss Ice Shelf and King George Island where stations tendto cluster in a small area from dominating the continental sta-tistics Snowfall represents a critical input to the continentrsquos icesheet mass balance but accurate continent-scale observationsare not available Therefore precipitation is not assessed inthis study

3 Results

[20] The problems associated with coarse resolution forAntarctic simulations were noted in section 24 Figure 1bshows the difference between model elevation and true stationelevation as a function of elevation and demonstrates theimpact of using higher model resolution At 60 km most ofthe stations have positive errors that can exceed 800m Whenthe resolution is increased to 15 km (red asterisks) the errorsare substantially reduced for stations below 1 km elevationAt both resolutions some stations that are on flat groundare still located on steeply sloping model terrain Forecastwind speed and direction errors from such misrepresentationscannot easily be corrected For temperature and pressure onthe other hand adjustments are made using the dry adiabaticlapse rate (~98 Ckm) and the hypsometric relationship

Figure 2 (andashc) Domain-averaged monthly 2m air temper-ature statistics lowest correlations are found in January(n= 25)


278

before any of the statistical analysis that follows The meanmodel bias (BIAS) root mean square error (RMSE) and cor-relation (CORR) statistics are computed with respect to thethree-hourly observations of air temperature dew point tem-perature surface pressure wind speed and direction and inci-dent surface shortwave and longwave radiation

31 Impact of Model Improvements on Forecast Skill

311 Surface Air and Dew Point Temperatures[21] The impact on forecast skill of model improvements

is assessed by comparing forecasts made using GFS-FNLfor a fixed period (January to December 2007) with threedifferent versions of the model (Polar WRF 301 311321 hereafter PWRF301 PWRF311 and PWRF321respectively) PWRF331 only became available after a largenumber of these simulations were completed and is there-fore evaluated just for January and July 2007 in the sensi-tivity experiments described in section 32 The model

physics improvements considered here are primarily inthe parameterizations of shortwave and longwave radiation(Table 1) Statistics are averaged over the stations in Figure 1afor each version of the model Figure 2 shows that the modelskill in forecasting temporal variability of 2m air temperatureas measured by the temporal correlations exceeds 07 formost of the year In January a lower value of ~055 indicatesdifficulties forecasting temporal variability at the peak ofsummer when the diurnal cycle is pronounced Spatialpatterns of the statistics (not shown) indicate higher correla-tions on the eastern plateau Although correlations vary littlebetween the model versions the biases show distinct differ-ences between the winter and the summer seasons All ver-sions of the model show a cold near-surface temperature biasin summer with January peaks below 2 C During winteron the other hand a warm bias prevails again for all threeversions The differences are greater in January (gt2Ccolder) compared with July when the warm bias remainslt2C Among the three versions PWRF321 generally shows

Figure 4 (andashc) Domain-averaged monthly station pressurestatistics (n = 46)

Figure 3 (andashc) Domain-averaged monthly 2m dew pointtemperature statistics (n = 51)


279

the smallest bias (both warm and cold) The RMSE differ-ences are also larger between the seasons (up to 10 C) thanthey are between model versions (lt05C) Stations withmore variable January temperatures also tend to exhibit thelargest cold bias Large variations in the statistics for adjacentstations (expected to show similar statistics) on the Ross IceShelf and near the South Pole underscore the need for strin-gent quality control of the observations The lowest correla-tions are found along the coast another indication of thedifficulty of accurately representing these stations In sum-mary changes in the model versions from 301 to 321 haveonly a small impact on the forecasting skill of the 2m air tem-perature over Antarctica[22] To assess forecasts of the moisture field we use

averages of monthly 2m dew point temperatures (Figure 3)The temporal patterns are very similar to those found in the2m air temperature above However the temporal correla-tions are slightly lower remaining near 07 throughout theyear as opposed to above 07 for air temperature Once againthere is little difference in the temporal correlation statistic

between model versions and hence the prediction skill oftemporal variability As with the 2m air temperature themodels have a cold bias in dew point temperatures inJanuary and a warm bias in July but the cold dew point biasin January is smaller (lt1C) than the corresponding air tem-perature bias (gt2C) Therefore the forecast January dewpoint depressions (temperature minus dew point temperatureat 2m) are smaller indicating a forecast atmosphere closer tosaturation than is observed In spite of this we show in sec-tion 313 that the model atmosphere has deficiencies in cloudcover that substantially impact the surface energy balanceWinter dew point temperature RMSEs are larger than thosefor temperature because dew point measurements at theselow temperatures are in general less reliable The very lowamounts of moisture on the continent together with difficul-ties in making the observation as well as in predicting theirvariations contribute to the larger RMSEs The differencesin seasonal correlation bias and RMSEs again greatly ex-ceed those between different model versions312 Surface Pressure and Wind Speed[23] Statistics for surface pressure are also very similar

between the different model versions and no single versionconsistently outperforms the others all year round The corre-lations of surface pressure are generally high (~09) exceptin January and May when they are about 08 (Figure 4)Although PWRF301 and PWRF311 show a smaller seasonaldifference in the bias a strong seasonal cycle is found in thePWRF321 bias which is also the lowest it remains near zeroin June and is negative (lt1 hPa) in July and October Fore-cast surface pressures are generally higher than observedleading to biases exceeding ~4 hPa in January and Decemberfor the two earlier versions The larger January positive pres-sure bias is consistent with the cold temperature bias we sawearlier[24] Correlations for forecast monthly wind speed exceed

~050 throughout the year with no well-defined differencebetween winter and summer (Figure 5) Values from the dif-ferent versions are again nearly equal indicating that all threepredict the temporal evolution of surface wind speed equallywell Unlike the correlations bias between forecast and ob-served wind speeds shows a strong seasonal cycle rangingin magnitude from ~1m s1 in February to ~25m s1 inAugust The peak winter wind speed bias likely is related toseasonal variation in absolute wind speeds with higherspeeds producing the larger biases As noted above the windfield is the most sensitive to model terrain and inaccuraterepresentation of complex topography cannot always be cor-rected easily Therefore in addition to the spatial averagesdescribed above statistics from a subset of stations in rela-tively homogeneous subregions for January 2007 are pre-sented in Table 2 Only PWRF321 results are shown becausethe others are very similar Consistent with Figure 5 averageforecast wind speeds are generally higher than observed andthe forecast wind direction generally falls within the samequadrant as the observed Also when grouped by subregions(low elevation ice shelf and interior) the least wind speedbiases clearly occur in the interior of the continent and thegreatest occur at the low elevations where they can exceed9m s1 both for average and for most frequent cases Theseresults therefore underscore the importance of choosingappropriate model resolutions especially for domains thatinclude the Antarctic coast where many of the verification

Figure 5 (andashc) Monthly mean wind speed statistics from25 stations


280

stations are located Overall PWRF321 which used theRRTMG radiation schemes has a smaller bias in tempera-ture dew point and pressure than the two earlier versionsof Polar WRF313 Downwelling Surface Shortwave and LongwaveRadiation[25] To determine the cause of the cold-summer and warm-

winter temperature biases discussed above we first examinethe surface radiation balance using the four BSRN stations atNeumayer Syowa Dome C and Amundsen-Scott The sitesrange in elevation from low-altitude coastal to high-altitudelocations on the interior of the East Antarctic Plateau Figure 6shows 2007 annual cycles of observed as well as forecastmonthly incident surface shortwave (SWDOWN) radiationfrom PWRF321 averaged over grid points that share similargeographic characteristics In January 2007 at Neumayer(a lower latitude station) the average forecast downwellingshortwave radiation does not exceed 360Wm2 whereas atAmundsen-Scott (higher latitude) about 400Wm2 is simu-lated (Figure 6c) These forecasts of higher January surface in-solation near the South Pole are consistent with observationsshowing ~380Wm2 incident at the surface at Amundsen-Scott compared with only ~310Wm2 at Neumayer To showthat the downwelling shortwave at Neumayer and Syowa isrepresentative of Antarctic coastal locations forecasts from10 additional sites that match the locations of establishedcoastal stations (Halley Dumont Casey Mirnyj MawsonRothera Belgrano Silvia Mt Siple and McMurdo) are usedThe forecast average from these sites is generally lower thanthe averages at Neumayer and Syowa suggesting that thereare some coastal locations where lower amounts of incidentsurface SWDOWN reach the surface compared with Neumayerand Syowa However the forecast average for the coastal

sites is still higher than the observations at Neumayer andSyowa from November to March when isolation plays alarger role in the surface energy budget Therefore on aver-age excessive forecasts of downwelling surface shortwaveradiation at Neumayer and Syowa are characteristic of Ant-arctic coastal locations A similar forecast excess is also pre-dicted at Dome C and South Pole but the differences aresmaller At Dome C the forecast average from five additionallocations on the East Antarctic Plateau (Vostok Dome ADome Fuji Kohonen and AGO-3) is generally lower thanobserved but the grid point closest to the station receivesmore downwelling shortwave radiation than observed Theaverage downwelling shortwave radiation around the SouthPole is calculated from sites matching the locations of HenryNico Panda South AGO-4 and Amundsen-Scott It is gen-erally higher than both forecast and observation for the SouthPole BSRN station The diurnal cycle (not shown) indicatesthat the large positive SWDOWN bias occurs mostly afterthe local noon especially at Neumayer A small-amplitudeforecast diurnal cycle is also found near the South Pole be-cause the pertinent model grid point is centered slightly offof the South Pole but even here the bias is positive (excess)throughout the day[26] To assess whether the positive bias in SWDOWN is

limited to 2007 a similar radiation analysis was performedfor 1993 Because the full annual simulations had beencompleted before the release of PWRF321 the results inFigure 7a are from PWRF311 instead The time series areof daytime and nighttime minimum from 1 July to 31December followed by 1 January to 31 July for both 1993and 2007 This rearrangement puts the peak downwellingsummer SWDOWN near the middle of the figure for clarityAlso note the lack of observed SWDOWN at Dome C in

Table 2 Surface Wind Speed Statistics from 3-Hourly Observations of a Representative Subset of Stations for January 2007a

Avg Speed Avg Direction Most Frequent Direction Most Frequent Speed

Latitude Longitudestddev OBS W321 OBS W321 OBS W321 OBS W321

n (N) (E) (ms) (ms) (ms) () () () () (ms) (ms)

Low ElevationRothera 239 676 681 3 4 9 51 54 22 45 6 9Syowa 238 690 396 3 4 5 75 81 67 90 5 7B San Martin 220 681 671 7 4 9 64 73 225 90 4 15Neumayer 244 707 83 4 7 7 105 96 90 90 9 10Mirnyj 120 666 930 3 5 14 143 104 135 90 6 15Casey 243 663 1105 5 5 11 82 99 180 90 4 13D-47 241 674 1387 5 12 15 154 130 157 135 13 17Arelis 248 767 1630 3 4 12 183 208 180 225 7 14Lola 140 741 1634 3 4 12 203 252 202 270 7 19Rita 248 747 1640 6 7 7 272 239 270 225 10 7

Ice ShelfVito 246 785 1778 2 4 6 196 184 180 180 5 6Lorne 151 783 1700 3 4 6 222 210 202 180 7 9Linda 247 785 1684 4 6 9 203 224 202 225 9 10Limbert 195 759 593 3 4 5 196 220 180 225 5 5Ferrell 245 779 1708 4 5 6 203 201 202 180 7 10

InteriorSky Blu 156 748 715 4 5 7 353 38 225 45 4 6Nico 246 890 897 4 4 6 308 324 337 0 5 6Amundsen 105 900 00 4 5 6 38 358 90 90 5 5Henry 246 890 10 2 4 6 38 70 157 90 4 6Concordia 184 751 1234 2 3 5 196 200 180 225 4 5

aThe forecasts from PWRF321 (W321) observed (OBS) speed direction and standard deviations (stddev) are computed from n observations in January2007 Average wind direction is obtained using mean zonal and meridional wind speeds The most frequent speeds and directions are based on a wind roseanalysis


281

1993 Figure 7a shows that the greater differences arefound in the maximum compared with the minimum Atboth Neumayer and Dome C the forecast minimum valuesin shortwave are very close to the observed The forecastmaximum SWDOWN on the other hand is close to theobservation at both Neumayer and South Pole only for largerand not smaller values of maximum SWDOWN The highsummer variability in surface downwelling shortwave atNeumayer is immediately apparent both in the observationsand in the forecasts but the observations clearly vary muchmore than the forecasts Variability in the forecast maximumSWDOWN decreases more rapidly than that in observationsduring the fall seasons and this can also contribute (greateraverages) to the excess SWDOWN during this part of the yearThus larger forecasts of the smaller maximum SWDOWNvalues and smaller variability at higher values are responsiblefor the forecasts of excessive downwelling shortwave radiation[27] To verify that cloudy skies are indeed a factor in

the SWDOWN bias forecasts of instantaneous clear-sky

downwelling shortwave radiation at South Pole were com-pared with those in which cloud interactions were allowedThe results (not shown) show a near-perfect match most ofthe time between the clear and the cloudy forecasts There-fore the excessive downwelling radiation results from a ten-dency in the model for a higher frequency of clear skiesFigure 7a also shows that the excessive SWDOWN occursin both years and is therefore not influenced by the interan-nual variations in the large-scale circulation Because themodel does predict downwelling shortwave better only underclear skies we hypothesize that inadequate cloud representa-tion (lower amounts or smaller optical thickness) is responsi-ble for the forecasts of excess surface incident shortwaveradiation[28] If the model clouds are indeed deficient a deficit in

the downwelling longwave (LWD) radiation should corre-spond to the excess in SWDOWN Figure 7b is a scatterplotof PWRF311 longwave down (y axis) versus the observed(x axis) In an ideal case the values would fall exactly alongthe diagonal (blue line) Poor agreement between forecastsand observations is indicated by displacement from thediagonal At Neumayer (top row) the LWD amounts clusterat both the low and the high amounts of observed LWDbut show a larger spread for intermediate amounts (150ndash280Wm2) during both 1993 and 2007 Observed LWDvalues for 1993 are missing for South Pole and Dome Cso only 2007 values appear in the bottom row They showclustering for lower values of LWD both at Dome C andat the South Pole More points fall below the diagonal atthe South Pole in 2007 and at Neumayer in 1993 and lessdistinctly in 2007 These results indicate an overall tendencyfor less LWD in the model than in the observations and areconsistent with the positive SWDOWN bias in Figure 6The pattern is not as striking at Dome C which also hadfewer measurements of longwave down[29] At South Pole and Dome C disparities between the

model and observations increase for higher LWD values(cloudy skies) further supporting a hypothesis of inadequatecloudiness The cloud deficit was investigated using ob-served cloudiness from NCDC for Neumayer Vostok andAmundsen-Scott (Figure 8a) Observed sky cover less than1 1ndash4 5ndash7 and above 7 oktas are usually classified as clear(CLR) scattered (SCT) broken (BKN) and overcast (OVC)respectively Although making cloud observations in Antarc-tica during the winter darkness is difficult the excess down-welling shortwave and deficient longwave radiation areprominent during the summer as well At Neumayer the fre-quency of overcast conditions is about 30 for all monthsand the combined frequency of overcast and broken sky con-ditions usually exceeds 60 This high frequency of cloudyconditions likely is due to the sitersquos proximity to the coastand frequent passages of synoptic storms and explains thelower mean SWDOWN at Neumayer (in spite of its lowerlatitude Figure 6) compared with South Pole where thecorresponding total is typically only 40 Observations fromVostok are used instead of those from Dome C because oflack of observed cloud categories at Dome C Clear skiesoccur more frequently at Vostok also on the plateau andwe infer a similar pattern for neighboring Dome C[30] Performing an objective cloud comparison remains dif-

ficult because models do not forecast cloud categories andbecause of uncertainties in the reported cloud observations that

Figure 6 (andashc) Mean monthly incident shortwave radia-tion at the surface for 2007 the average is computed forlocations of similar characteristics from the forecasts SyowaBSRN observations are available only for January


282

depend on the observerrsquos ambient weather conditionsDiffering definitions of model total cloud fraction and empir-ical algorithms used to infer cloud cover from satellitemeasurements also make it difficult to compare model cloudfraction directly against satellite measurements RecentCloud-Aerosol Lidar and Infrared Pathfinder Satellite Obser-vations (CALIPSO) provide profiles of cloud liquid waterpath that can be more objectively compared with forecastsof the same fields from the model and therefore motivatefuture study

[31] In spite of these difficulties forecast total cloud frac-tion can be estimated from the atmospheric water contentThe formulation used here combines model cloud liquidand ice water paths as modified for use in Antarctica by Fogtand Bromwich [2008] It shows spatial patterns with highercloud fractions at Neumayer and lower amounts toward theSouth Pole in qualitative agreement with the observationsdescribed above In Figure 8b the derived cloud estimate isoverlaid on the observed category (represented using themidrange fraction in oktas) in January 2007 Observed

Figure 7 (a) Forecast (red) and observed (black) daytime (max) and nighttime (min) observed surfacedownwelling radiation (Wmndash2) at (top row) Neumayer (middle row) Dome C and (bottom row) SouthPole only the maximum downwelling shortwave radiation is shown for South Pole because of smalldiurnal amplitude Left column is for 1993 and the right column is the corresponding radiation for2007 The gap in the x-axis indicates the discontinuity resulting from rearrangement of the time serieswhich places summer insolation in the middle of the plots (b) Scatterplots of PWRF311 forecast surfacedownwelling longwave radiation (Wmndash2) for (top row) Neumayer 1993 and 2007 The bottom row is forSouth Pole and Dome C in 2007 The blue line marks the one-to-one line representing a perfect fit


283

cloudier conditions (gray bars) at Neumayer relative to theother two sites are immediately obvious and so is the fact thatthe computed total cloud fraction remains near zero for muchof the month Higher model total cloud fraction (red dashes)also corresponds to observed (gray bars) overcast conditions(7580) most of the time For intermediate observed cloudi-ness (3080 and 6080) the model cloud fraction is alwaysclose to zero In other words the model is able to forecastovercast events better than the broken and scattered sky condi-tions Table 3 shows results from both PWRF311 in 1993 and2007 as well as PWRF321 just for 2007 and therefore allowsan examination of the impact of interannual differences inthe surface radiation balance in addition to model improve-ments The SWDOWN statistics are computed only from thesunlit part of the year (1 September to 1 March) because ofthe strong seasonal cycle and many zero values during theAntarctic winter LWD statistics on the other hand are com-puted for the entire year because of the critical role LWD playsin the surface energy balance during the winter months Exceptfor South Pole in 1993 the temporal correlations of downwel-ling shortwave radiation exceed 09 at all three locations inboth years and model versions It is interesting that the corre-lation at Neumayer where the SWDOWN variability is largeis comparable to that at both Dome C and South Pole

Correlations for downwelling longwave are more variableand range from 060 to 080 and are lower than those ofSWDOWN This suggests that temporal variation downwel-ling shortwave radiation which is dominated by the diurnalcycle is better forecasted than that in LWD which comes pri-marily from clouds in the atmosphere The biases show anexcess in SWDOWN at all locations but greatest at the SouthPole The excess also occurs in both years as well as in the twomodel versions analyzed but the bias in LWD is mixed withDome C showing more than observed surface longwave radi-ation for PWRF311 and PWRF321 The larger RMSE in bothdownwelling short and longwave radiation at Neumayer is notsurprising given the high variability there We conclude thatthe model has difficulty predicting the downward longwaveradiation and hence the surface energy balance because ofdifficulties in correctly predicting cloud cover and its effects314 Upper Air Variables[32] Figure 9 shows forecast and observed temperature

profiles averaged from two observed profiles daily (at 0000and 1200 UTC) available from upper air stations in the IGRAarchive therefore the three-hourly forecasts are resampled tomatch the observation frequency and at the standard levels(850 700 500 400 300 250 and 200 hPa) Table 4 givesresults for all variables at 12 stations (see Figure 1a for

Figure 7 (continued)


284

locations) Only the forecasts from PWRF311 are shownbecause small variations occur between the model versionsin the upper air statistics and for consistency with the radia-tion analysis described above In general differences fromobservations in the upper air variables are smaller than atthe surface Both observed and predicted profiles clearlyshow the tropopause (~300 hPa) during summer but themodel overestimates its temperature The model forecaststhe broad characteristics of the temperature profile accuratelyin summer and winter even capturing the large annual tem-perature range above 200 hPa (~60K) which is nearly threetimes than that near the surface

[33] During the summer (December January and February[DJF]) the temperature correlations range from 079 to 096whereas biases remain generally cold and small except at thetropopause where they are warmer than observed by at least1C (Table 4) In winter the temperature correlations nearthe bottom of the profile are slightly higher than in summerUnlike the surface biases discussed above upper air temper-ature biases are small and remain less than 1C except at thetropopause A cold DJF bias occurs near the bottom of theprofile in all versions and is consistent with the surface tem-perature bias discussed earlier The RMSE values indicatethat winter temperatures are more variable especially below

Figure 8 (a) Observed cloud frequencies from NCDC for clear (CLR) scattered (SCT) broken (BKN)and overcast (OVC) categories in 2007 at Neumayer Vostok and Amundsen-Scott stations (b) Derivedcloud fraction from observations (gray) and forecasts (red dashed) for January 2007 Cloudy conditionsare reported at Neumayer more frequently all year and the forecasts underestimate the total cloud coverat all three locations


285

500hPa but in summer a greater variability occurs near thetropopause Correlations between forecasts and observed geo-potential heights are generally high and remain above 09 inboth seasons In all three versions geopotential height biasesare smaller near the bottom but increase with height to at leastthree times as large at 200 hPa as they are at 700 hPa Some re-cent work [Cassano et al 2011 Powers et al 2010] suggeststhat spectral nudging near the model top or different formula-tion of the model layer top could reduce this error In fact cor-recting this anomaly was another motivation for using theRRTMG radiation scheme but this was not successful[34] Away from the complex Antarctic topography the

upper air meridional and zonal wind speeds show highercorrelations than 08 in both seasons above 700 hPa Anom-alous positive wind speed bias above 700 hPa is foundprimarily in January for the zonal component but the zonalwind speed bias is slightly larger (absolutely) than that of themeridional component At 850 hPa the zonal wind is weakerthan observed whereas the meridional is stronger but theabsolute bias is smaller than that found at the surface

Table 3 Correlations Biases and RMSE at 3 h Intervals forDownwelling Shortwave and Longwave Radiationa

Shortwave Down Longwave Down

PWRF311 PWRF321 PWRF311 PWRF321

1993 2007 2007 1993 2007 2007

CorrelationNeumayer 094 095 094 065 077 076Dome C mdash 098 098 mdash 079 080South Pole 070 097 097 mdash 063 060

Bias (Wm2)Neumayer 478 328 275 188 94 129Dome C mdash 199 80 mdash 62 58South Pole 533 660 542 mdash 234 238

RMSE (Wm2)Neumayer 1080 887 909 406 358 363Dome C mdash 463 511 mdash 213 197South Pole 738 582 468 mdash 292 304

aStatistics for downwelling shortwave radiation are for the periods21 September to 31 December and 1 January to 21 March Statisticsfor downwelling longwave radiation are for the entire year

Figure 9 Observed and forecast mean vertical temperature profiles from 2007 for six Antarctic stationsfrom IGRA black and gray dots are forecasts and observed temperatures respectively blue curves areaverages for the winter months (JunendashAugust) and the red curves represent summer (DecemberndashFebruary)temperatures Observed temperatures are continuous curves forecasts are represented by dashed curvesNote that observations at 1000 hPa are intermittently available for Neumayer Mirnyj McMurdo Mawsonand Casey and at 700 hPa for Amundsen-Scott


286

Statistics for the total wind speed (Table 4) show similarcharacteristics higher winter correlations and marked heightdependence in both seasons Relative humidity statisticsshow the lowest skill of all the variables Only values forlevels below 400 hPa are shown because the measurementsare less reliable with many missing observations higher upThe highest summer and winter correlations occur below700 hPa and are 065 and 074 respectively Biases are posi-tive in summer and mixed in winter and are generally lessthan 5

32 Sensitivity to Physics Schemes InterannualVariability and Resolution

[35] This section discusses experiments in which the modelsensitivity to interannual variations radiation physics PBLschemes uncertainty in driving data and horizontal modelresolution are evaluated Only January and July are analyzedbecause they showed the largest seasonal contrast Improve-ment in the observation network in Antarctica is evident fromoccurrence of more than twice as many useable observationsin 2007 as in 1993 The uncertainty for each month is calcu-lated by the ratio of the standard deviation to the square rootof the number of stations used from that month An evalua-tion of the latest version of Polar WRF (331) is includedin Table 5 (column 7) This version was released after manyof the experiments described above had been completed[36] The first two columns in Table 5 show statistics from

the two different years (1993 and 2007) for the same config-uration of PWRF311 driven by the same analysis (ERA-Interim) The summer temperature correlation RMSE andbias are all higher in 2007 than in 1993 but the aforemen-tioned cold bias is exhibited in both years A similar patternis seen in the surface pressure and wind speed statistics butnot in dew point for which the correlations RMSE and biasare higher in 1993 than in 2007 A cold summer and strongerwind speed bias are common to both Polar WRF forecasts

During July the surface temperature bias is cold in these runsdriven by ERA-Interim in sharp contrast to the warm biasfound in the earlier experiments driven by GFS-FNL Thisfinding indicates that the model is very sensitive to the dataused to provide the lateral boundary and initial conditions[37] The next three columns in Table 5 assess the sensitivity

of PWRF321 to radiation and PBL physics so the GFS-FNLdata are used in all three experiments Temperature correla-tions are comparable and the biases are negative in summerand positive in July in agreement with the full-year experi-ments that also used GFS-FNL described in section 31 Thecorrelations are also higher during July for all three The runwith the community atmospheric model (CAM) radiation hasthe least winter 2m temperature bias Dew point temperaturesshow the least summer skill in all three experiments but thewinter skill is slightly better than the corresponding summervalues Wind speed statistics show that all three experimentshave stronger than observed wind speeds in both months andthat for each month the individual statistics from the experi-ments are comparable Replacing the MYJ PBL scheme inPWRF321 with the MYNN PBL [Nakanishi and Niino 2004]scheme (columns 3 and 5 in Table 5) increases the summercold bias but does not substantially change the other statis-tics The final two columns in Table 5 compare the newestversion of Polar WRF (331) with the preceding version(321) In these two experiments the driving data and phys-ics combinations are exactly the same[38] The most striking feature when the results from experi-

ments with ERA-Interim are compared with those with GFS-FNL is the drastic reduction in both the summer and the wintertemperature biases to near zero A possible explanation is thatthe GFS-FNL analysis initializes Polar WRF to a state that istoo cold for the model to recover from within a 48 h periodA similar improvement is found in the wind speed statisticsin which a positive bias still exists but is noticeably smallerTo compare objectively all the experiments from 2 years

Table 4 Domain-Averaged Upper Air Statistics for DecemberndashFebruary Representing Summer and JunendashAugust for Winter (n= 12)

DJF JJA DJF JJA

hPa BIAS RMSE CORR BIAS RMSE CORR BIAS RMSE CORR BIAS RMSE CORR

Air temperature (K) Polar WRF311 Geopotential heights (m) Polar WRF311850 031 154 086 093 231 090 16 154 096 74 176 099700 005 154 088 034 201 088 23 160 092 94 216 094500 037 141 093 055 150 093 33 211 097 113 251 098400 012 159 090 101 160 093 12 244 097 77 271 098300 152 236 079 046 139 090 109 308 097 45 297 098250 034 240 089 016 159 091 246 383 098 108 319 098200 063 168 096 068 184 095 444 518 098 314 414 099Zonal wind speed (ms) Polar WRF311 Meridional wind speed (ms) Polar WRF311850 021 386 079 136 516 083 101 368 065 106 475 063700 058 364 081 011 413 080 021 319 075 171 394 080500 077 370 084 061 429 090 013 368 087 019 423 089400 080 441 086 059 491 090 007 476 087 007 485 090300 018 515 086 035 530 090 009 570 088 007 520 091250 034 408 088 059 522 092 005 444 091 021 447 092200 081 286 090 196 532 093 089 289 093 029 365 093Wind speed (ms) Polar WRF311 Relative humidity () Polar WRF311850 036 402 080 163 538 078 171 176 065 172 189 074700 046 334 072 011 403 074 181 191 061 085 202 072500 028 423 079 044 408 083 674 188 062 348 200 065400 045 548 079 042 469 084 757 175 052 536 198 062300 209 433 081 068 515 087250 138 250 085 095 477 089200 005 198 092 046 526 089


287

with different numbers of observations and disparate physicscombinations and driving data rank scores are computed Cor-relations for each variable are ranked across the experimentsfrom the largest to the smallest and corresponding RMSEand bias from the smallest to the largest individual ranks arethen summed to define the rank scores shown at the bottomof Table 5 PWRF331 and PWRF311 have the best overallperformance in January and July respectively but PWRF321does well in both months[39] Section 3 has shown vertical displacement of the

location of some stations in the model as a result of coarseresolution An additional sensitivity experiment was con-ducted to examine how the model skill would be impactedby resolution higher than 60 km The experiment in the col-umn under PWRF331 in Table 5 was repeated at the 15 kmresolution used by AMPS for the Antarctic continent Table 6compares statistics for low-elevation stations (below 1 km)where elevation errors were largest Except for the fourfoldincrease in horizontal resolution the two sensitivity experi-ments are otherwise identical For 2m air temperaturethe January correlation decreases from 066 to 064 and thebias decreases from 124 to 062K During July on the

other hand correlations are in general lower at 15 km Thispossibly is due to strong synoptic influence in July whichmay not benefit from the higher resolution Statistics for thesurface pressure show the least sensitivity to the increasedmodel resolution in both months The forecast statistics fordew point temperature show higher skill in the 15 km run inboth January and July having higher correlations and smallerbias and RMSE The largest sensitivity to model resolution isfound in the forecast skill for January wind speeds The corre-lation increases whereas both the RMSE and the bias de-crease In fact the wind speed bias is reduced by 50 or morein both January and July The results show that although highresolution is beneficial even higher resolution may be neededto capture fully the complex coastal environment[40] Scatterplots shown in Figure 10 compare experiments

with PWRF321 (columns 4 and 5 of Table 5) that use theRRTMG and CAM radiation and MYJ and MYNN PBLphysics respectively Figure 10 allows for an evaluation ofmodel sensitivity in areas without observed data in theSouthern Ocean and over Antarctica and also depicts therange of model-to-model difference over the entire domainThe top row shows the impact of analysis (ERA-Interim

Table 5 January and July Correlations (CORR) RMSE and Bias From the Sensitivity Experiments (n=Number of Stations)a

ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim

January2m Temp (n= 23 65)CORR 057 (003) 063 (003) 061 (002) 061 (002) 059 (002) 061 (002) 062 (002)RMSE (K) 359 (047) 411 (019) 355 (017) 385 (019) 470 (025) 353 (017) 341 (017)BIAS (K) 183 (055) 305 (026) 128 (026) 212 (026) 332 (030) 009 (030) 006 (029)

Stn Pressure (n= 17 68)CORR 092 (002) 097 (001) 096 (001) 093 (001) 096 (001) 097 (001) 095 (001)RMSE (hPa) 360 (060) 423 (026) 468 (028) 545 (028) 464 (028) 425 (024) 498 (025)BIAS (hPa) 014 (066) 057 (045) 001 (048) 008 (047) 006 (047) 073 (041) 133 (044)

2m Dew Point (n= 17 30)CORR 055 (005) 051 (004) 046 (004) 043 (004) 051 (004) 044 (004) 044 (004)RMSE (K) 545 (111) 357 (024) 473 (032) 445 (029) 403 (025) 552 (040) 541 (039)BIAS (K) 203 (082) 130 (039) 293 (040) 216 (041) 115 (043) 403 (046) 392 (046)

Wind Speed (n= 7 18)CORR 024 (004) 060 (004) 055 (005) 054 (004) 055 (005) 059 (005) 061 (004)RMSE (ms) 437 (080) 353 (044) 367 (041) 370 (043) 368 (044) 343 (041) 335 (038)BIAS (ms) 069 (118) 160 (038) 140 (039) 152 (038) 164 (039) 107 (039) 086 (0039)Rank score 68 43 54 59 52 42 33

July2m Temp (n= 26 67)CORR 077 (003) 078 (001) 077 (001) 077 (002) 077 (001) 078 (002) 075 (002)RMSE (K) 623 (149) 589 (024) 565 (024) 556 (018) 570 (024) 565 (024) 572 (027)BIAS (K) 137 (080) 101 (044) 169 (037) 011 (041) 187 (037) 089 (038) 017 (040)

Stn Pressure (n= 1865)CORR 095 (00) 097 (001) 097 (001) 094 (001) 097 (003) 097 (001) 095 (001)RMSE (hPa) 306 (084) 408 (023) 451 (023) 526 (024) 447 (023) 410 (021) 485 (022)BIAS (hPa) 037 (082) 091 (041) 043 (044) 036 (042) 036 (043) 108 (037) 167 (040)



aPhysics combinations used and whether the experiment was driven by GFS-FNL or ERA-Interim lateral boundary conditions (LBC) are also indicatedEnclosed in parentheses are the estimated ranges of uncertainty in the sample statistics The uncertainty for each month is calculated by the ratio of the stan-dard deviation to the square root of the number of stations used from that month The rank score is the sum of the ranks (high correlation low rank high biasand RMSE low ranks) for each experiment from each variable High overall skill corresponds to low rank score


288

versus GFS-FNL) and the bottom row depicts model sensi-tivity to the radiation physics (CAMMYJ versus RRTMGMYNN) because these two PBL schemes produce very sim-ilar results (Table 5) The differences (y axis) are calculatedat each model grid point for January 2007 and then plottedagainst the value (x axis) for that grid point from PWRF321forecasts of 2m temperature surface pressure wind speedand downwelling shortwave radiation (Figures 10andash10d re-spectively) More locations are colder in the GFS-FNL runthan in the ERA-Interim reanalysis run for temperatures

below 270K More locations in the CAMMYJ experimentare warmer (reduced the colder summer bias) than in theRRTMG experiment for temperatures in the 230ndash270Krange In both cases the temperature differences becomesmaller for temperatures above 274K The surface pressurescatterplot (Figure 10b) shows larger surface differences near1000 hPa (over the ocean primarily) and smaller valuescorresponding to lower pressures (over elevated Antarctica)Below 850 hPa forecast surface pressures from the GFS-FNL run are generally lower than those from the ERA-Interim experiment This could result from differences inthe locations of synoptic systems between the two analysesin these experiments For CAMMYJ versus RRTMGMYNN experiments which both use GFS-FNL the surfacepressure differences are almost symmetrical about zero(bottom row in Figure 10b) Consistent with statistics shownabove the wind speeds are generally stronger in the GFS-FNL experiments relative to the ERA-Interim experiment(Figure 10c) Figure 10 also shows that between the analy-sis- and the reanalysis-driven experiments positive differ-ences are more frequent for lower to intermediate windspeeds but negative for speeds exceeding 10m s1 Weakerwinds (less than 5m s1) are over predicted especially inthe GFS-FNL run Switching the radiation physics does notcause as much spread in the wind speed as does switchingthe driving analysis[41] Figure 10d shows that the GFS-FNL run forecasts more

surface downwelling shortwave radiation than the ERA-Interim for values less than 400Wm2 Switching the radia-tion physics to CAM reduces this bias but has only a modestimpact for shortwave radiation exceeding 600Wm2Figure 10 also shows that above 800Wm2 differencesbetween the experiments can approach 600Wm2 Thus thechanges in incident surface shortwave radiation resulting

Table 6 Average Statistics for Low-Elevation (lt1 km) Stations at60 and 15 km for January and July 2007a

PWRF331 January 2007 July 2007

Resolution 60 km 15 km 60 km 15 km

2m Temp (n= 35 30)CORR 066 064 081 074RMSE (K) 329 322 553 603BIAS (K) 124 062 059 022

Stn Pressure (n= 33 33)CORR 097 097 097 097RMSE (hPa) 238 222 384 331BIAS (hPa) 053 062 017 018

2m Dew Point (n= 20 21)CORR 056 058 080 084RMSE (K) 309 303 654 556BIAS (K) 053 059 104 038

Wind Speed (n= 28 24)CORR 065 067 058 058RMSE (ms) 304 281 524 453BIAS (ms) 095 020 252 130

aExcept for the fourfold increase in horizontal resolution the two sensitivityexperiments are identical The two numbers enclosed in parenthesis are thenumber of stations used for January and July At each resolution exactly thesame stations are used but not all have observations in both months

Figure 10 January 2007 scatterplots of Polar WRF 321 differences with (top row) GFS-FNL minusERA-Interim runs and (bottom row) GFS-FNL with CAMminus GFS-FNL with RRTMG The scatterplotsshow the range of differences found at Polar WRF grid points between two sensitivity experiments for(a) 2m temperature (b) surface pressure (c) wind speed and (d) downwelling surface shortwave radiation


289

from a change in physics have a smaller (smaller differences)impact on the forecast SWDOWN compared with changingthe driving analysis (top row in Figure 10d) From this analysiswe conclude that Polar WRF is more sensitive where the 2mair temperatures are below 270K (over the ice sheet) andwhere surface pressures are above 950 hPa (coastal regionsand Southern Ocean) and that the model is more sensitiveto the driving analysis than a change in radiation schemeThis suggests that although comparisons with observationsdiscussed above yield average model statistics that are compa-rable between different model versions the model is sensitiveto the conditions over the Southern Ocean where observationsare scarce and a poor representation of the synoptic systemscould be detrimental to the forecast over the ice sheet

4 Discussion

[42] The main findings can be summarized as follows Theforecast skill of Polar WRF improved only slightly betweenversions 301 and 331 Forecast 2m temperatures exhibita cold summer and warm winter bias and the wind speeds

are stronger than observed A deficit in the downwellinglongwave resulting from inadequate cloud representation isinferred to be responsible for the summer cold bias Qualita-tive support for this hypothesis is found when observed andestimated cloudiness are compared The excess in downwel-ling shortwave radiation plays a secondary role because~80 is reflected back and lost to space with the high surfacealbedo and clear model skies in Antarctica To demonstratethat deficit in downwelling longwave radiation explains mostof the summer surface temperature bias we use the surfaceenergy budget equation Assuming a balance in which thenet shortwave radiation sensible latent and snowpack heatflux all remain unchanged the temperature change thatresults from increasing the downward longwave radiationbased on the deficits from Table 3 is calculated using theStefan Boltzmann equation The results show an increasein summer surface temperatures of 14 and 29K at Neumayerand South Pole respectively An increase of 2K representsmore than 60 of the cold bias (Table 5a) in the GFS-FNLruns[43] Recall that the summer cold bias is almost eliminated

when ERA-Interim reanalysis is used in the sensitivityexperiments described in section 32 This indicates that theanalysis used is also an important factor in the cold bias andmost likely is communicated through the input moisture fieldERA-Interim sensitivity experiments with a smaller cold biasshow higher moisture flux into the atmosphere and hencemore atmospheric water vapor a well-known greenhousegas across the continent Many radiation schemes currentlyin use rely on relative humidity cloud water or cloud ice todiagnose cloud-radiation interaction yet we have found rela-tive humidity to be one of the fields predicted with the leastskill[44] Surface energy fluxes at locations corresponding to

the BSRN sites are shown in Figure 11 as are the skin tem-peratures and the temperatures at 8m depth The 8m temper-ature remains constant throughout the year and is equal tothe annual mean for 2007 However the skin temperatureshows both diurnal and seasonal variations During summerthe skin temperature is generally warmer than at 8m depthresulting in a reversal of the snowpack temperature gradientbetween the seasons As a result of our treatment of the snow-pack temperature profiles (section 23) the heat fluxes withinthe snowpack are small and play only a secondary role in thetemperature biases discussed above Without this treatmenthowever an extremely sharp temperature gradient developsin the top layer of the snowpack and enhances the downwardheat flux producing a fairly rapid temperature increase atdepths below 2m within the 48 h integration period[45] The net shortwave radiation fluxes are positive dur-

ing summer whereas the net longwave radiation is negative(energy loss) all year round Figure 11 shows a net radiativeheating (net shortwave + net longwave radiation) at Neu-mayer The ground heat flux exceeds 10Wm2 at Dome Cand South Pole but not at Neumayer Both the latent andthe snowpack heat fluxes are small in July so the balanceduring this time is between the fluxes of downward sensibleheat and net longwave radiation loss from the surface Thewarm winter bias discussed above must therefore come pri-marily from excessive downward sensible heat fluxes causedby too much mixing in the stable Antarctic boundary layerassociated with the positive wind speed bias Stronger than

Figure 11 PWRF321 annual cycle of 2007 daily averagednet shortwave (SW red) and longwave (LW blue) radiationsurface sensible (SHFLX blackasterisk) latent (LHFLXgreen) and ground (GRDFLXmagenta) heat fluxes at the threeBSRN sites The sign convention for SHFLX and LHFLX ispositive upward into the atmosphere GRDFLX is negative ifdownward from the surface The scale at right is for the skin(maroon) and deep snowpack (yellow) temperatures


290

observed wind speeds appear to be a common problem forstable boundary layers with all PBL schemes used here aswell as with the Yonsei PBL (C Mass personal communica-tion 2011) The sensible heat flux along the coast can be sub-stantially larger than on the plateau as a result of mechani-cally induced mixing supported by the anomalously strongforecast winds We hypothesize that the model winter fluxesof sensible heat are larger than they should be but observedestimates in Antarctica vary too much to be useful in evaluat-ing these simulations Town and Walden [2009] compiledestimates of sensible and latent heat fluxes at or near theSouth Pole Observed July sensible heat fluxes (negativedownwards) range from about5 to 25Wm2 The modelpredicted sensible heat flux directed toward the surface fromthe sensitivity runs range from 208 to 298Wm2 and istherefore close to the highest observational estimate[46] The results described here have a winter bias (warm)

similar to that of the shorter integrations of Tastula andVihma [2011] who used a different analysis (ERA-40 re-analysis) and year (1998) in their comparison Using 28 sitesin an evaluation of Polar MM5 forced with ECMWFTOGA Guo et al [2003] found an average cold bias of~2C throughout the year together with positive pressurewind speed and mixing ratio biases and proposed yearlongseries of short integrations to evaluate the model furtherSuch integrations have been completed in this study andshow comparable statistics Thus we can conclude thatthe statistics presented here are representative of the skillof Polar WRF and that the temperature and wind speedbiases identified are a recurrent problem in Antarctic fore-casts These must be addressed because they affect the sim-ulated snowdrift during the cold months and ice melt duringthe melt season and therefore the Polar WRF simulated sur-face mass balance It is important for these factors to besimulated accurately for reliable prediction of long-termchanges in ice sheet surface mass balance

5 Conclusions

[47] We have examined three factors that can influence theskill of Polar WRF in Antarctic forecasts The results showthat recent versions of Polar WRF have skill that is compara-ble to that of a recent Arctic simulation [Wilson et al 2011]and marginally better than that of an earlier Antarctic simula-tion with Polar MM5 [Guo et al 2003] The model skill var-ies seasonally reflecting the different roles played by localand synoptic scale systems at different times of the yearThe model shows a cold surface temperature bias in the sum-mer months but a warm bias during winter months when in-solation plays a negligible role in heating the surface Themodel over predicts the summer incident surface shortwaveand under predicts the longwave down at the surface Fore-cast skill is better away from the complex topography Thelower skill near the surface is due partially to the 60 km reso-lution used here but the summer cold bias is driven primarilyby deficits in downwelling longwave radiation No nudgingwas done in this study although it has been suggested[Cassano et al 2011] that spectral nudging near the modeltop can improve the performance of Polar WRF The statis-tics presented here therefore provide a pure forecast bench-mark of the skill of Polar WRF in Antarctic simulationsComparison of the statistics from different years physics

schemes and analyses shows that the model skill is affectedmore by the analysis used than by the parameterization phys-ics or year-to-year differences in the atmospheric circulationThus improving the network of verification observations andquality of Antarctic analyses must remain a top priority forfurther development of numerical modeling in Antarctica

[48] Acknowledgments This work was funded by NASA grantNNX08AN57G and AMPS grants NSF ANT-1135171 NSF OPP-0838967 and NSF ANT-1049089 The authors appreciate the support ofthe Antarctic Meteorological Research Center for the providing the AWSdata (Matthew Lazzara and Elena Willmot NSF grant ANT-0838834)Aaron Wilson provided highly appreciated and valuable discussions fromhis insight from working with Polar WRF in the Arctic

ReferencesBracegirdle T J and G J Marshall (2012) The reliability of Antarctictropospheric pressure and temperature in the latest global reanalysesJ Clim 25 7138ndash7146 doihttpdxdoiorg101175JCLI-D-11-006851

Bromwich D H and Z Liu (1996) An observational study of the katabaticwind confluence zone near Siple Coast West Antarctica Mon WeatherRev 124 462ndash477

Bromwich D H J J Cassano T Klein G Heinemann K M HinesK Steffen and J E Box (2001) Mesoscale modeling of katabatic windsover Greenland with the Polar MM5Mon Weather Rev 129 2290ndash2309

Bromwich D H K M Hines and L-S Bai (2009) Development andtesting of Polar WRF 2 Arctic Ocean J Geophys Res 114 D08122doi1010292008JD010300

Bromwich D H D F Steinhoff I Simmonds K Keay and R L Fogt(2011a) Climatological aspects of cyclogenesis near Adelie Land AntarcticaTellus 63A(5) 921ndash938 doi10111j1600-0870201100537x

Bromwich D H J P Nicolas and A J Monaghan (2011b) An assessmentof precipitation changes over Antarctica and the Southern Ocean since1989 in contemporary global reanalyses J Clim 24 4189ndash4209doi1011752011JCLI40741

Cassano J J and T R Parish (2000) An analysis of the nonhydrostaticdynamics in numerically simulated Antarctic katabatic flows J Atmos Sci57 891ndash898 doi1011751520-0469(2000)057lt0891AAOTNDgt20CO2

Cassano J J M E Higgins and M W Seefeldt (2011) Performance ofthe Weather Research and Forecasting (WRF) Model for month-longPan-Arctic simulations Mon Weather Rev 139 3469ndash3488 doi101175MWR-D-10-050651

Chen F and J Dudhia (2001) Coupling an advanced land-surfacehydrol-ogy model with the Penn StateNCAR MM5 modeling system Part IModel implementation and sensitivity Mon Weather Rev 129 569ndash585doi1011751520ndash0493(2001)129lt0569 CAALSHgt20CO2

Chini G P and S Leibovich (2003) An analysis of the Klemp andDurran radiation boundary condition as applied to dissipative internalwaves J Phys Oceanogr 33 2394ndash2407 doi1011751520-0485(2003)033lt2394AAOTKAgt20CO2

Comiso J (1999 updated 2007) Bootstrap sea ice concentrations fromNIMBUS-7 SMMR and DMSP SSMI [2007] Boulder ColoradoNational Snow and Ice Data Center Digital media

Dee D P et al (2011) The ERA-Interim reanalysis configuration andperformance of the data assimilation system Q J R Meteorol Soc137(656) 553ndash597 doi101002qj828

Durre I R S Vose and D B Wuertz (2006) Overview of the IntegratedGlobal Radiosonde Archive J Clim 19 53ndash68 doi101175JCLI35941

Durre I R S Vose and D B Wuertz (2008) Robust automated qualityassurance of radiosonde temperatures J Appl Meteorol Climatol 472081ndash2095 doi1011752008JAMC18091

Dutton E G (2008) Basic and other measurements of radiation at stationSouth Pole (2007-05) Climate Monitoring amp Diagnostics LaboratoryBoulder doi101594PANGAEA706267

EPICA Community Members (2004) Eight glacial cycles from an Antarcticice core Nature 429 623ndash628 doi101038nature02599

Fogt R L and D H Bromwich (2008) Atmospheric moisture and cloudcover characteristics forecast by AMPS Weather Forecast 23 914ndash930doi1011752008 WAF20061001

Gemmill W B Katz and X Li (2007) Daily real-time global sea surfacetemperaturemdashHigh resolution analysis at NOAANCEP NOAANWSNCEPMMAB Office Note 260 39 pp NOAA Silver Spring Md

Guo Z D H Bromwich and J J Cassano (2003) Evaluation of PolarMM5 simulations of Antarctic atmospheric circulation Mon Wea Rev131 384ndash411 doi1011751520-0493(2003)131lt0384EOPMSOgt20CO2


291

Hannon J (2012) Op Deep Freeze 2011ndash2012 season concluded AmericasNorth shore journal httpnorthshorejournalorgop-deep-freeze-2011-2012-season-concluded

Hines K M and D H Bromwich (2008) Development and testing of PolarWRF Part I Greenland Ice Sheet meteorology Mon Weather Rev 1361971ndash1989 doi1011752007MWR21121

Hines K M D H Bromwich L-S Bai M Barlage and A G Slater(2011) Development and testing of Polar WRF Part III Arctic landJ Clim 24 26ndash48 doi1011752010JCLI34601

IAATO (2010) Antarctic Tourism fact sheet httpiaatoorgtourism-statistics

Koumlnig-Langlo G (2011) Basic and other measurements of radiationat Neumayer Station (2007-03) Alfred Wegener Institute for Polar andMarine Research Bremerhaven doi101594PANGAEA759355

Laine V (2008) Antarctic ice sheet and sea ice regional albedo and temper-ature change 1981ndash2000 from AVHRR Polar Pathfinder data RemoteSensing Environ 112 646ndash667

Liu H K Jezek B Li and Z Zhao (2001) Radarsat Antarctic MappingProject digital elevation model version 2 Boulder CO National Snowand Ice Data Center Digital media

Mlawer E J S J Taubman P D Brown M J Iacono and S A Clough(1997) RRTM a validated correlated-k model for the longwave J Geo-phys Res 102 16663ndash16682

Monaghan A J D H Bromwich H Wei A M Cayette J G PowersY-H Kuo and M Lazzara (2003) Performance of weather forecastmodels in the rescue of Dr Ronald Shemenski from South Pole in April2001 Weather Forecast 18 142ndash160

Monaghan A J D H Bromwich J G Powers and KW Manning(2005) The climate of the McMurdo Antarctica region as representedby one year of forecasts from the Antarctic Mesoscale Prediction SystemJ Clim 18 1174ndash1189

Nakanishi M and H Niino (2004) An improved Mellor-Yamada level-3model with condensation physics Its design and verification BoundaryLayer Meteorol 112 1ndash31

National Research Council (2011) Future science opportunities in Antarcticaand the Southern Ocean httpwwwscribdcomdoc64382324Future-Science-Opportunities-in-Antarctica-and-the-Southern-Oceanarchive

NCAR (1999) Data set ds0832 NCEP FNL operational model global tro-pospheric analyses httpdssucaredudatasetsds0832 Boulder Colo

Nicolas J P and D H Bromwich (2011) Precipitation changes in highsouthern latitudes from global reanalyses A cautionary tale Surveys inGeophysics 32 475ndash494 doi101007s10712-011-9114-6

Nigro M A J J Cassano M A Lazzara and L M Keller (2012) Casestudy of a barrier wind corner jet off the coast of the Prince Olav Moun-tains Antarctica Mon Weather Rev 140 2044ndash2063

NSF (2012) Antarctic medical evacuation Press release 12-149Ohmura A and Coauthors (1998) Baseline Surface Radiation Network(BSRNWRMC) a new precision radiometry for climate research BullAm Meteorol Soc 79 2115ndash2136

Parish T R and D H Bromwich (2007) Re-examination of the near-surface air flowover theAntarctic continent and implications on atmosphericcirculations at high southern latitudesMon Weather Rev 135 1961ndash1973

Powers J G (2007) Numerical prediction of an Antarctic Severe WindEvent with the Weather Research and Forecasting (WRF) model MonWeather Rev 135 3134ndash3157

Powers J G S M Cavallo and K W Manning (2010) Improving upper-level performance in AMPS Longwave radiation 5th Antarctic Meteoro-logical Observation Modeling and Forecasting Workshop 12ndash14 July2010 BPRC Columbus OH

Powers J K W Manning D H Bromwich J J Cassano and A M Cayette(2012) A decade of Antarctic science support through AMPS Bull AmMeteorol Soc 93 1699ndash1712

Reynolds R W T M Smith C Liu D B Chelton K S Casey andM G Schlax (2007) Daily high-resolution blended analyses for seasurface temperature J Clim 20 5473ndash5496 doi1011752007JCLI18241

Rignot E I Velicogna M R van den Broeke A Monaghan andJ Lenaerts (2011) Acceleration of the contribution of the Greenlandand Antarctic ice sheets to sea level rise J Geophys Res 38doi1010292011GL046583

Seefeldt M W and J J Cassano (2012) A description of the Ross IceShelf air stream (RAS) through the use of self-organizing maps (SOMs)J Geophys Res 117 D09112 doi1010292011JD016857

Skamarock W C J B Klemp J Dudhia D O Gill D M Barker X-YHuang WWang and J G Powers (2009) A description of the AdvancedResearch WRF Version 3 NCAR Technical Note NCARTNndash475+ STR113 pp

Steinhoff D F (2011) Dynamics and variability of Foehn winds inthe McMurdo Dry Valleys Antarctica PhD dissertation Atmo-spheric Sciences Program The Ohio State University ColumbusOh 271 pp

Steinhoff D F D H Bromwich and A J Monaghan (2012) Dynamics ofthe foehn mechanism in the McMurdo Dry Valleys of Antarctica from Po-lar WRF Quart J Roy Meteor Soc doi 101002qj2038

Tastula E-M and T Vihma (2011) WRF Model experiments on the Ant-arctic atmosphere in winter Mon Weather Rev 139 1279ndash1291doi1011752010MWR34781

Town M S and V P Walden (2009) Surface energy budget over theSouth Pole and turbulent heat fluxes as a function of an empirical bulkRichardson number J Geophys Res 114 D22107 doi1010292009JD011888

Velicogna I (2009) Increasing rates of ice mass loss from the Greenlandand Antarctic ice sheets revealed by GRACE Geophys Res Lett 36L19503 doi1010292009GL040222

Vitale V (2009) Basic measurements of radiation at Concordia station(2006-02) Institute of Atmospheric Sciences and Climate of the ItalianNational Research Council Bologna doi101594PANGAEA711756

Weller G and P Schwerdtfeger (1977) Thermal properties and heat trans-fer processes of low-temperature snow Antarctic meteorological studiesat plateau station Antarctica Antarctic Res Ser 25 27ndash39

Wilson A B D H Bromwich and K M Hines (2011) Evaluation ofPolar WRF forecasts on the Arctic System Reanalysis domain I Surfaceand upper air analysis J Geophys Res 116 D11112 doi1010292010JD015013

Wilson A B D H Bromwich and K M Hines (2012) Evaluation ofPolar WRF forecasts on the Arctic System Reanalysis domain 2Atmospheric hydrologic cycle J Geophys Res 17 D04107doi1010292011JD016765

Yamanouchi T (2010) Basic and other measurements of radiation at sta-tion Syowa (2007-04) National Institute of Polar Research Tokyodoi101594PANGAEA740932


292

and in potential large-scale evacuations in the future In addi-tion archived AMPS forecasts have been used extensively tostudy Antarctic weather and climate [eg Monaghan et al2003 2005 Fogt and Bromwich 2008 Powers 2007Powers et al 2010 2012 Steinhoff 2011 Bromwich et al2011a Nigro et al 2012 Seefeldt and Cassano 2012Steinhoff et al 2012][4] Model simulations produce data that in the absence of

long-term observations can be used to diagnose climaticprocesses such as those depicted in Antarctic ice corescovering eight glacial cycles [European Project for Ice Cor-ing in Antarctica (EPICA) 2004] Such applications willrequire climate models that accurately predict the short-termweather (forecasts) before they can be used to diagnoselong-term climate process such the accelerated rate of iceloss from the continent [Velicogna 2009 Rignot et al2011] Obtaining a complete understanding of Antarcticteleconnections and climate variability using models is anarea of active research and a recent report of the NationalResearch Council (NRC) [2011] supports the developmentof a new generation of robust earth system models[5] The skill of Polar WRF already demonstrated in the

Arctic may not prevail in Antarctica because of substantialdifferences between the two polar regions First the atmo-spheric circulation in Antarctica is strongly influenced bythe presence of the large ice sheet with elevations exceeding3 km in most areas Any mesoscale model used here mustaccurately represent the strong katabatic winds that aregoverned by the balance of gravity thermal stability andsynoptic forcing [eg Bromwich and Liu 1996 Cassanoand Parish 2000 Parish and Bromwich 2007] The coldtemperatures limit evaporation and sublimation and makeice the dominant phase of water in Antarctic clouds Highsurface albedo [Laine 2008] and relatively pristine skies(low amounts of aerosol) also result in different radiation-atmosphere interactions Surrounded by a large oceanAntarctica has limited observational data needed to initializemesoscale models and to validate forecasts making the veri-fication of models more difficult than in the Arctic[6] Development of the standard WRF is a community

effort led by the National Center for Atmospheric Research(NCAR) among others Benefits of community involvementin the development include rapid and continuous refinementsto both the standard and the Polar WRF The primary draw-back to such rapid developments is that newer model ver-sions are subject to fewer evaluations than their predecessorsAll past verifications cannot be repeated every time a newmodel is released Incompatibilities in the physics parameter-izations and other errors (bugs) also arise occasionally fromthe model developments themselves In a recent Arctic-wideevaluation Polar WRF 311 demonstrated significant Arcticforecast skill This study expands the evaluations to all ofAntarctica on an annual time scale and is the sixth in a series[Hines and Bromwich 2008 Bromwich et al 2009 Hineset al 2011Wilson et al 2011 2012] of verification studiesdocumenting ongoing Polar WRF developments We focuson three factors that influence forecast skill model improve-ments uncertainty in driving data and interannual varia-tions in the large-scale atmospheric circulation The impactof model improvements is assessed using four recent ver-sions of Polar WRF (releases 301 311 321 and 331)Bug-fixed subversions are chosen because they include

corrections for problems identified by the user communitysince the initial release Improvements that only make themodel easy to use enhance computational efficiency orspeed up diagnosis of model output are not considered[7] This study examines the impact on forecast skill of un-

certainty in the initial and lateral boundary conditions bycomparing Polar WRF forecasts made using the reanalysisfrom the European Centre for Medium-Range Weather Fore-casts (ECMWF ERA-Interim) [Dee et al 2011] and thefinal analysis from the National Centers for EnvironmentalPrediction (NCEP) operational global analysis (GFS-FNL)The relative accuracy of contemporary reanalyses has beenexamined previously in relation to predicted precipitationand ERA-Interim fared the best [Nicolas and Bromwich2011 Bromwich et al 2011b] In addition from a compar-ison of five reanalyses with Antarctic station observations[Bracegirdle and Marshall 2012] ERA-Interim was foundto be the most reliable for mean sea level pressure and500 hPa height trends ERA-Interim represents a majoradvance over previous ECMWF reanalyses because ofincreased resolution (T255) a more advanced atmosphericmodel 4D-Var assimilation system and the enormous vol-ume of assimilated data Assimilation of satellite radiancedata is especially critical over the Southern Ocean wherethere are few synoptic stations However reanalyses suchas ERA-Interim have the advantage of a fixed modeling sys-tem and input data that are usually unavailable in operationalforecasting Thus many centers rely on operational analysesfor lateral boundary conditions For this reason this studycompares Polar WRF skill when driven by an analysis anda reanalysis The two cover a reasonable range that can beassociated with the uncertainty in the lateral boundary dataand are considered sufficient for the objectives here How-ever we note that many other sources of lateral and initialconditions exist[8] Influence of interannual variations in the large-scale

atmospheric circulation on model skill is evaluated by com-paring forecast statistics from two different years (1993 and2007) both made using the same analysis (ERA-Interim)We choose 1993 to allow for a comparison between thestatistics given here with those of an earlier Polar MM5 Ant-arctic verification [Guo et al 2003] and using 2007 allowsus to take advantage of recent advances in the observingnetwork Strictly speaking the quality of a reanalysisdepends on the number of assimilated observations andthese did differ between 1993 and 2007 thereby compound-ing differences resulting from interannual variations withthose resulting from analysis quality However weassume that ERA-Interim representations of Antarctic atmo-spheric circulation are among the best possible for these years[Bromwich et al 2011b]

2 Model Experiments Lower BoundaryConditions and Observations

21 Model Description

[9] Polar WRF [Hines and Bromwich 2008 Hines et al2011] was developed by the Polar Meteorology Group atThe Ohio State University and builds on previous successwith an earlier mesoscale model (Polar MM5) [Bromwichet al 2001] in the Arctic The overall difference is that PolarWRF is modified with enhancements that adapt the standard


275
















276









277






3 Results




278









279








280








n (N) (E) (ms) (ms) (ms) () () () () (ms) (ms)






281









282





283






284





285






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292
















276









277






3 Results




278









279








280








n (N) (E) (ms) (ms) (ms) () () () () (ms) (ms)






281









282





283






284





285






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292









277






3 Results




278









279








280








n (N) (E) (ms) (ms) (ms) () () () () (ms) (ms)






281









282





283






284





285






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292






3 Results




278









279








280








n (N) (E) (ms) (ms) (ms) () () () () (ms) (ms)






281









282





283






284





285






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292









279








280








n (N) (E) (ms) (ms) (ms) () () () () (ms) (ms)






281









282





283






284





285






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292








280








n (N) (E) (ms) (ms) (ms) () () () () (ms) (ms)






281









282





283






284





285






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292








n (N) (E) (ms) (ms) (ms) () () () () (ms) (ms)






281









282





283






284





285






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292









282





283






284





285






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292





283






284





285






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292






284





285






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292





285






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292






1993 2007 2007 1993 2007 2007







286









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292









DJF JJA DJF JJA




287






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292






ModelYearPWRF311

1993PWRF311

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF321

2007PWRF331

2007PhysicsAnalysis

RTMMYJERA-Interim

RTMMYJERA-Interim

RRTMGMYJGFS-FNL

CAMMYJGFS-FNL

RRTMGMYNGFS-FNL

RRTMGMYNERA-Interim

RRTMGMYNERA-Interim











288














289


4 Discussion








290



5 Conclusions
























291




































292














289


4 Discussion








290



5 Conclusions
























291




































292


4 Discussion








290



5 Conclusions
























291




































292



5 Conclusions
























291




































292




































292

comprehensive evaluation of polar weather research and...

Documents