Romanian Journal of Meteorology, Vol. 7, No. 1, 2005 ISSN 1223-1118 TESTING OF THE NON-HYDROSTATIC MODEL LM (LOKAL MODELL) ON THE ROMANIAN TERRITORY Liliana VELEA, I. V. PESCARU, Rodica DUMITRACHE National Meteorollogical Administration (NMA), Bucharest, Romania I. K. MOUSTAFA University of Bucharest, Romania (Manuscript received 15 April 2005, in final form 22 August 2005) Abstract: The nonhydrostatic model LM was tested for two horizontal resolutions, of 14 km and of 2.8 km, for 2-5 November 2004 and January 2005. The results have been analyzed, at 14 km resolution, by comparison with the hydrostatic model HRM at the same resolution and with observed data from 16 stations. At 2.8 km, the model outputs have been compared with hourly data from the measurement campaign on 2-5 November 2004 at Baia Mare. The model accurately describes the interaction of non-hydrostatic processes with large-scale processes. The analysis of results shows an improvement of the model forecast accuracy, especially for precipitation. Key works: non-hydrostatic, high-resolution, limited-area model 1. INTRODUCTION Most of the current numerical weather prediction models operate on hydrostatic scales of motion with a grid spacing of about 10 km and thus lack the spatial resolution required to explicitly capture some small-scale severe weather events. By employing a 1 to 3 km grid spacing, it is expected that deep moist convection and the associated feedback mechanism to the larger scales of motion can be solved explicitly. In addition, the impact of topography on the organization of penetrative convection, e.g. by channeling effects, is represented much more realistically in high resolution nonhydro- static forecast models. The nonhydrostatic Lokal Modell (LM) has been developed in its basic version at Deutscher Wetterdienst (DWD), Germany, the subsequent developments being organized within COSMO (Consortium for Small Scale Modelling). It has been designed for meso-ȕ and meso-Ȗ scales where nonhydrostatic effects begin to play an essential role in the evolution of the atmospheric flows. But the physical representation still needs fine tuning of the involved parameters in order to obtain a better forecast on the desired domain. Also the optimum combination of numerical schemes available at different horizontal scales needs to be found. The LM model has been used in a series of experiments on the Romanian territory at two horizontal resolutions, 14 km and 2.8 km, in order to test its ability to simulate the medium and small scale phenomena and to highlight the improvements in the numerical prediction of near-surface weather conditions that it can provide. A short description of the model is given in section 2 and the experiment settings are presented in section 3. The model results from experiments at the resolution of 14 km are
Transcript
Romanian Journal of Meteorology, Vol. 7, No. 1, 2005 ISSN 1223-1118
TESTING OF THE NON-HYDROSTATIC MODEL LM (LOKALMODELL) ON THE ROMANIAN TERRITORY
Liliana VELEA, I. V. PESCARU, Rodica DUMITRACHENational Meteorollogical Administration (NMA), Bucharest, Romania
I. K. MOUSTAFAUniversity of Bucharest, Romania
(Manuscript received 15 April 2005, in final form 22 August 2005)
Abstract: The nonhydrostatic model LM was tested for two horizontal resolutions, of 14 km and of2.8 km, for 2-5 November 2004 and January 2005. The results have been analyzed, at 14 kmresolution, by comparison with the hydrostatic model HRM at the same resolution and with observeddata from 16 stations. At 2.8 km, the model outputs have been compared with hourly data from themeasurement campaign on 2-5 November 2004 at Baia Mare. The model accurately describes theinteraction of non-hydrostatic processes with large-scale processes. The analysis of results showsan improvement of the model forecast accuracy, especially for precipitation.
Key works: non-hydrostatic, high-resolution, limited-area model
1. INTRODUCTION
Most of the current numerical weatherprediction models operate on hydrostaticscales of motion with a grid spacing ofabout 10 km and thus lack the spatialresolution required to explicitly capturesome small-scale severe weather events.By employing a 1 to 3 km grid spacing,it is expected that deep moist convectionand the associated feedback mechanismto the larger scales of motion can besolved explicitly. In addition, the impactof topography on the organization ofpenetrative convection, e.g. by channelingeffects, is represented much morerealistically in high resolution nonhydro-static forecast models.
The nonhydrostatic Lokal Modell(LM) has been developed in its basicversion at Deutscher Wetterdienst(DWD), Germany, the subsequentdevelopments being organized withinCOSMO (Consortium for Small Scale
Modelling). It has been designed formeso- and meso- scales wherenonhydrostatic effects begin to play anessential role in the evolution of theatmospheric flows. But the physicalrepresentation still needs fine tuning ofthe involved parameters in order to obtaina better forecast on the desired domain.Also the optimum combination ofnumerical schemes available at differenthorizontal scales needs to be found.
The LM model has been used in aseries of experiments on the Romanianterritory at two horizontal resolutions,14 km and 2.8 km, in order to test itsability to simulate the medium and smallscale phenomena and to highlight theimprovements in the numerical predictionof near-surface weather conditions that itcan provide. A short description of themodel is given in section 2 and theexperiment settings are presented insection 3. The model results fromexperiments at the resolution of 14 km are
L. VELEA, I.V. PESCARU, R. DUMITRACHE, I.K.MOUSTAFA
14
compared with the results of thehydrostatic model HRM (High resolutionRegional Model). The two models wereintegrated on the same domain and at thesame resolutions and the results ofsimulations are verified against synopticdata from 16 meteorological stations. Theverification method is similar to theoperational verification procedure used bythe National Meteorological Admini-stration (NMA) Bucharest for productsprovided to users in the aeronautical field.For the resolution of 2.8 km, the resultsare qualitatively compared with observeddata from a measuring campaign on2-5 November 2004 at Baia Mare. Theanalysis of results is presented insection 4.
2. DESCRIPTION OF THE MODELS
The Lokal Modell is a limited-areaatmospheric prediction model, based onthe nonhydrostatic, full compressiblehydro-thermodynamical equation inadvection form. The model equations areformulated in rotated geographicalcoordinates, using an Arakawa C-grid,and a generalized terrain-following heightcoordinates, with a Lorenz vertical gridstaggering. The time integration is doneusing a second-order leapfrog HE-VI(horizontally explicit, vertically implicit)time-split integration scheme, includingextensions proposed by Skamarock andKlemp (1992). There are also availableoptions for a two-time level secondorder Runge-Kutta split explicit scheme(Wicker and Skamarock, 1998) and for athree time-level 3-d implicit scheme(Thomas et al., 2000). For the numericalsmoothing a 4th-order linear horizontaldiffusion scheme with monotonicorographic limiter is used (Doms, 2001).
The LM radiative transfer package isbased on the -two stream radiationscheme introduced by Ritter and Geleyn
(1992) for short and longwave fluxes,with full cloud-radiation feedback. Itemploys eight spectral intervals andincorporates the effects of scattering,absorption and emission by clouddroplets, aerosols and gases (H2O, CO2,O2, O3, N2O, CH4) in each part of thespectrum.
For determining grid-scale cloudsand precipitation, the cloud watercondensation and evaporation arecomputed by the saturation adjustmentprocess: if a grid box becomessupersaturated during a time step, thetemperature and the concentrations ofwater vapor and cloud water areisobarically adjusted to a saturated state,taking into account the latent heat. Ifcloud water is present in spite ofsubsaturation, it is evaporated until eitherno cloud water remains or saturation isachieved. The saturation is watersaturation, as cloud ice is not dealt with.Precipitation formation is treated by abulk microphysics parameterizationincluding water vapor, cloud water, rainand snow with column equilibrium for theprecipitating phases. Also, options for abulk scheme including cloud ice andfor 3-d precipitation transport areavailable. The subgrid scale cloudiness isinterpreted by an empirical functiondepending on the relative humidity andheight. Corresponding cloud water contentis also quantified.
There are two available schemes formoist convection treatment: the mass-fluxconvection scheme (Tiedtke, 1989) withequilibrium closure based on moistureconvergence, and a Kain-Fritsch (1993)convection scheme with non-equilibriumCAPE-type closure. At very finehorizontal resolutions, convection istreated explicitly.
The vertical turbulent transportbetween the atmospheric layers is treatedusing a 2-nd order turbulent closure with aprognostic equation for the turbulent
TESTING OF THE NON-HYDROSTATIC MODEL LM ON THE ROMANIAN TERRITORY15
kinetic energy (TKE). Compared to theformer parameterization scheme whichis based on diagnostic second orderK-closure, this has the advantage ofcontaining additional terms that describemore physical processes (i.e. subgrid-scalecondensation, thermal circulation, verticalentrainment), which affect the evolutionof TKE in time.
Turbulent transport through thetransfer layer between the rigid surfaceof the earth and the lowest atmosphericmain level is parameterized using ascheme based also on the prognostic TKEequation, which does not use empiricalprofile functions, as it is done in theformer scheme (Louis, 1979). The transferlayer consists of two parts, a roughnesslayer and a Prandtl layer above it.Different interpolation formulae for theturbulent velocity scale in both layers areused and a laminar sub-layer isconsidered, just above the rigid surface,where only laminar diffusion takesplace. This transfer scheme leads tosignificantly improved mean profiles ofthe 2m-temperature and especially the2m-dewpoint temperature, with realisticamplitudes of the diurnal cycle.
The soil processes are treated usingsoil model TERRA (Schrodin et al.,1995), which provides the surfacetemperature and the specific humidity atthe ground. The ground temperature iscalculated by the equation of heatconduction, which is solved in anoptimized two-layer model using theextended force-restore method (Jacobsenand Heise, 1982). The soil water contentis predicted for two, three or more layersby the Richards equation. Evaporationfrom bare land surfaces together withtranspiration by plants are derived asfunctions of water content and – only fortranspiration – of radiation and ambienttemperature. Most parameters of the soilmodel (heat capacity, water storagecapacity, etc) strongly depend on soil
texture. Five different types aredistinguished: sand, sandy loam, loam,loamy clay and clay. Additionally, threespecial soil types are considered: ice, rockand peat. Hydrological processes in theground are not considered for ice androck. However, potential evaporation isassumed to occur over ice, where the soilwater content remains unchanged.
There is also available a newmultilayer soil and vegetationmodel-TERRA_LM (Doms and Schattler,2001), where the effect of freezing/melting of soil water/ice is included,the process of snow melting is changedand a time-dependent snow albedo isintroduced. The new multi-layer conceptavoids the dependence of layer thicknesson soil type. Additionally it avoids theuse of different layer structures for thethermal and the hydrological sections ofthe model.
The equations for the hydrostaticmodel HRM (High resolution RegionalModel) are formulated in a rotatedlatitude/longitude grid using a hybridvertical coordinate (Majewski, 2005). Asplit semi-implicit time-stepping scheme,according to (Burridge, 1975) is used fortime integration and a linear 4th orderhorizontal diffusion scheme, with a slopecorrection for temperature is used fornumerical smoothing. Some of thephysical parameterizations of HRM arethe same as in LM: the radiative transferscheme, the mass-flux convection schemeafter Tiedtke (1989), the soil modelTERRA, while others are the same as informer schemes employed in LM:similarity theory-based surface transferscheme (Louis, 1979), level-2 scheme(Mellor and Yamada, 1974) of verticaldiffusion in the atmosphere.
3. EXPERIMENT DESIGN
The LM model has been tested for twohorizontal resolutions. At the resolution of
L. VELEA, I.V. PESCARU, R. DUMITRACHE, I.K.MOUSTAFA
16
14 km, the domain size is aproximati-vely 1000 x 900 km, with 35 verticallayers, and covers the Romanian territory(fig. 1a). The domain used for theresolution of 2.8 km is 100x100 km insize, with 50 vertical layers and it iscentered on Baia Mare (fig. 1b).
The global model GME outputs, at40 km resolution and interpolated on themodel grid, were used as drivingfields, with a frequency of updating thelateral boundary conditions of 3 hours,using a one-way nesting, Davies-typelateral boundary formulation. The terrain
a)
b)
Fig. 1. Topography of the integration domains: a) for LM and HRM models, 14 km horizontalresolution; b) for LM model, 2.8 km horizontal resolution. Units are m. Contour intervals are 200m
for a) and 100m for b).
TESTING OF THE NON-HYDROSTATIC MODEL LM ON THE ROMANIAN TERRITORY17
and surface data, containing all theexternal parameters needed by the LM andHRM models, were available for eachmodel resolution. The initial and boundaryconditions for experiments at 2.8 km areinterpolated from LM outputs at 14 km.All experiments have been carried out ona Linux cluster.
The simulation periods are January2005 and 2-5 November 2004, for theresolution of 14 km, and 2-5 November2004 for the resolution of 2.8 km. In allexperiments 24h forecasts were provided.
The results of LM integration at theresolution of 14 km are analyzed bycomparison with HRM, integrated at thesame resolution and on the same domain.Also, a statistical verification of theresults from both models against synopticdata was done. The verification methodapplied follows the requests of specificusers in the aeronautical field and also thestandards applied operationally at NMAThe 16 meteorological stations (table 1)are located on Romanian airports and theparameters considered for verification arethe following: sea level pressure (SLP),air temperature at 2m (T2m), wind speed
at 10m and precipitation cumulated inintervals of 6h. In order to obtain theforecast of parameters for all consideredstations, the method of the closestneighboring grid point has been used.
For the first three parameters thefollowing verification measures have beenused: mean error, standard deviation(STD) and mean square root error(RMSE). The precipitation modelforecasts have been analyzed using thefollowing scores: frequency bias, falsealarm ratio (FAR), probability of detection(POD), percent correct (PC), true skillscore (TSS), threat score (critical successindex, (TS). Frequency bias measures themodel’s ability to forecast events at thesame frequency as found in the sample,disregarding forecast accuracy. FalseAlarm Ratio is calculated as a ratiobetween the number of an event’s non-occurrences and the total number of thatevent’s forecasts, being therefore sensitiveto false predictions of the event, not tomissed events. Probability of Detection isa score that measures the ability toforecast correctly a certain category, thusbeing sensitive to missed events, not to
Table 1. Details of synoptic stations used for statistical verification of models at the14 km resolution.
L. VELEA, I.V. PESCARU, R. DUMITRACHE, I.K.MOUSTAFA
18
false alarms, and in an ideal situation ithas the value of one. Percent Correct is ameasure of forecast accuracy, calculatedas the ratio between the number ofcorrect forecasts and the total number offorecasts, expressed in percentages. Amore complete indicator is the TrueSkill Score, that uses all the relevantinformation contained in the observationand forecast, allowing an estimation ofthe probability that the observation/forecast association is real and its value isin the fixed range –1 to +1. The last scoreused for the verification of precipitation isThreat Score, which is a measure ofrelative accuracy and has the advantagethat it is sensitive to both false alarms andmissed events (Stanski et al., 1989).
For the resolution of 2.8 km, theresults from LM integration are analyzedonly by comparison with the observationdata obtained during the measurementcampaign on 2-5 November 2004 at BaiaMare. The campaign was realized withinthe project AIRFORALL (Air PollutionForecasting, Alert and Monitoring Systemon Short Time Scale, at local and regionalscale, in unfavorable meteorological andtopographic conditions). The purpose ofthe campaign was to make meteorologicaland pollution measurements in somelocations from Baia Mare and itssurroundings, in order to validate thenumerical forecast databases of bothmeteorological fields and pollution levels.There were eight fixed measurementpoints during the campaign: fourmanual measurements and four automaticstations. The principal reason in choosingthe measurement points was the represen-tativeness for local topography processesat fine scale and as a mean of thegrid-size mesh of the model used in theproject (about 3 km). The followingmeteorological parameters weremeasured: wind direction and intensity at2m (or 30m for some automatic stations);air temperature at 2m; relative humidity
at 2m; atmospheric surface pressure;global radiation (only for automaticstations); net radiation (only for automaticstations). A meteorological database wascreated, with a frequency of one hour, foreach day of the campaign.
4. RESULTS
a) LM simulations for the horizontalresolution of 14 km
In the following, some examples ofcomparisons between the LM model andthe HRM model, for November 2, 2004are presented. The meteorologicalsituation of this date was characterizedby a quite high temperature over thecountry for this period, with a maximumof 23.3 oC measured at Baia Mare, lowwind and light, sparse precipitation.
The spatial distribution of 2m-temperature difference between the twomodels, LM and HRM is shown infigure 2. It can be seen that therepresentation of nonhydrostatic effects inLM leads to a higher forecast accuracy (ofthe forecast) for this parameter in regionswhere the nonhydrostatic processesdominate the minimum represented scale(e.g. intra-Carpathian region). Deviationsof T2m values up to 2-3 degrees incomparison with the measured data, forthis situation, are observed.
Figure 3 shows the vectorialdifference in the wind field superimposedon sea level pressure difference betweenLM and HRM, and in figure 4 the plotsfor vertical velocities at levels of 850 hPaand 1000 hPa are presented. SLP in LMis lower in regions affected by verticalmass transport (south-western, northernpart of the country, figure 4b) andinteractions of non-hydrostatic processeswith large scale processes. The non-hydrostatic tendency in vertical velocitygives a higher accuracy of wind speed at10m in comparison with the HRM results,
TESTING OF THE NON-HYDROSTATIC MODEL LM ON THE ROMANIAN TERRITORY19
although still overestimated. Differencesin SLP are up to 2 hPa in an interval of24h, generally lower for LM if comparedwith the measured data for this simulation
period, and the wind speed correction isup to 3 m/s for this case.
Looking at the total precipitation in24h (fig. 5 a, b), a significant reduction
Fig. 2. Difference in T2m (shaded, units oC) and in SLP (contours, units hPa) between LM and HRMat 14 km resolution, for 02.11.2004, 24 h anticipation. Contour intervals are 2 oC for T2m and 0.5 hPa
for SLP.
Fig. 3. Difference with respect to SLP (shaded, contour interval 0.5 hPa) and wind vector (arrows),between LM and HRM at 14 km resolution, for 02.11.2004, 24 h anticipation.
L. VELEA, I.V. PESCARU, R. DUMITRACHE, I.K.MOUSTAFA
20
both in quantity and spatial distribution ofprecipitation in LM simulations can benoted, but this parameter is stillsignificantly overestimated.
A statistical verification of bothmodels against synoptic data has also beendone for January 2005. The scores for2m-temperature, sea level pressure and
a) b)
c) d)
Fig. 4. Vertical velocity (Pa/s) for 02.11.2004, 24 h anticipation, in : a) HRM, at level of 850 hPa; b) LM,at level 850 h Pa; c) HRM, at level 1000 hPa; d) LM, at level 1000 hPa. Contour interval is 0.02 Pa/s.
a) b)Fig. 5. Total precipitation in 24h interval for 02.11.2004, at 14 km resolution: a) for HRM; b) for LM.
Units are mm/day.
TESTING OF THE NON-HYDROSTATIC MODEL LM ON THE ROMANIAN TERRITORY21
wind speed at 10m have been computedfor each station and each anticipation(6h, 12h, 18h and 24h). Figures 6-9 showa monthly synthesis of forecast accuracyfor the two models, performed for the 16stations and two anticipations (18h and24h) allowing the possibility to follow theevolution of forecast quality betweenanticipations and between stations.
Analyzing the plots in figure 6 (a-c),an improvement in the forecast quality ofthe LM model in comparison with theHRM model can be observed. There isalso a similitude in the evolution of thetwo models’ forecast, since both modelsunderestimate the forecast value of SLP,as shown by the negative bias for allstations but Bucuresti-Baneasa (fig. 6a).For most stations, the mean error is lessthen 2 mb, excepting Caransebes station,where it doubles for both models. The plotfor standard deviation (fig. 5b) shows analmost uniform distribution of theforecast-observation differences aroundthe mean error, for both models and for allanticipations. It can be seen that for theanticipation of 18h the standard deviationhas a maximum value of 6, but for theanticipation of 24h the variance doublesand the maximum is now of 19.
If for SLP the mean errors aresystematically negative, in the case of 2m-temperature (fig. 7, a-c) the deviation signdiffers between stations. The bias is in theorder ±1.5 oC for both models and bothanticipations, excepting again Caransebesstation, where the mean error is about -3oC. The STD values show that the spreadof the instantaneous error is differentbetween stations and also betweenmodels: for LM, the STD is less then 8,for the anticipation of 18h, and has amaximum of 12 for the anticipation of24h, while for HRM there are stationswhere STD is higher then 20, showinga high frequency of the cases whenthe instantaneous error is far from themean.
a)
b)
c)
Fig. 6. Skill scores for sea level pressure, forJanuary 2005: a) mean error (mb); b) standard
deviation; c) mean square root error.
For wind speed at 10m, the errors werecomputed only for those cases when theobserved values were at least 4 m/s. This
L. VELEA, I.V. PESCARU, R. DUMITRACHE, I.K.MOUSTAFA
22
a)
b)
c)
Fig. 7. Skill scores for 2m-temperature, forJanuary 2005: a) mean error; b) standard
deviation; c) mean square root error.
condition is similar to the standard used inthe NMA’s operational verification ofthis parameter, where the winds withspeed less then this threshold are not
subject to statistical verification. Figures 8(a, b) indicate a quite good forecastaccuracy of both models for thisparameter for certain stations(i.e. Timisoara and Bucuresti-Baneasa),where STD is less than 2, for bothmodels, but there are also stations(i.e. Cluj and Tulcea) where high valuesof the scores (STD about 4, RMSE about3) were obtained.
a)
b)
Fig. 8. Skill scores for wind speed at 10m, forJanuary 2005: a) standard deviation; b) mean
square root error.
For precipitation, the verificationprocedure is applied for the Producing/NonProducing category and there isno reference to the forecasted/observedquantity of precipitation. From theanalysis of the scores computed for this
TESTING OF THE NON-HYDROSTATIC MODEL LM ON THE ROMANIAN TERRITORY23
parameter (fig. 9 a-f), it can be seen thatthe hydrostatic model HRM overestimatesthe producing of precipitation, thereforehaving a quite high probability ofdetection. In the LM case, the high PCvalues of the forecast, in agreement withthe low FAR, emphasize a better forecastquality of this model in comparison withthe HRM forecast.
The statistical evaluation of the LMand HRM models for these fourparameters emphasizes a better quality ofthe LM forecast. However, these arepreliminary conclusions as the sampledata set is too short and a more detailedanalysis is necessary for cases whensignificant errors were obtained.
a) b)
c) d)
e) f)
Fig. 9. Scores for precipitation cumulated for 6h interval, for January 2005: a) frequency bias; b) percentcorrect; c) probability of detection; d) false alarm ratio; e) true skill score; f) threat score.
L. VELEA, I.V. PESCARU, R. DUMITRACHE, I.K.MOUSTAFA
24
b) LM simulations for thehorizontal resolution of 2.8km
The results of a qualitativecomparison between the LM modelsimulations at 2.8 km resolution andthe measured data are presented below,for 3 November 2004. The meteorologicalsituation of this date was characterized, aswell as in the previous day, by a quitehigh air temperature, with a maximum of22 oC at Baia Mare. There was lightprecipitation on large areas in the south ofthe country, but no precipitation in theBaia Mare region.
For the considered date, the localsub-diurnal cycle of 2m-temperature isaccurately reproduced at fine resolution aswell as in the simulation at 14 km (fig.10),but the values for both resolutions aresystematically underestimated.
Fig. 10. Diurnal variation of 2m-temperature (oC)at Baia Mare, for 03.11. 2004; blue –observation;
red- LM at 2.8 km resolution; green- LMat 14 km resolution.
The local rotation of wind vector,for wind at 10m (fig. 11a-d) is wellrepresented at this resolution and it is ingood agreement with hourly measureddata. This feature is also present in LMsimulations at 14 km resolution, but lessemphasized and it does not appear in thehydrostatic model HRM.
The absolute values of forecastsurface fields show a systematicdifference when compared withobservations, which suggests that a fine
tuning of the parameters in physicalrepresentations (e.g. roughness length)could significantly improve thequantitative forecast of these fields.
5. CONCLUSIONS
The nonhydrostatic LM model has beentested for two horizontal resolutions: of14 km and of 2.8 km on domainscovering the Romanian territory and,respectively, centered on Baia Mare.
The comparison with the hydrostaticmodel HRM at 14 km resolution showsthat the nonhydrostatic processes consi-dered in LM, as well as the improvedparameterizations and numerical schemes,lead to a higher forecast accuracy of theanalyzed parameters (SLP, 2m-temperature, wind speed at 10m,precipitation). This is also found in theresults of both models’ statisticalevaluation against the synoptic data from16 stations. By analyzing these scores, itcan be seen that the LM model showsan improvement of forecast quality incomparison with the HRM forecast forall considered parameters and allanticipations. However, for both modelsthe forecast quality is diminished forlong-range anticipations. It should benoted that, as the mean values werecomputed for a short dataset, of only30 days, the results of models’ statisticalevaluation should not be generalized.Especially for precipitation, the scoresshould be computed for a longer periodof time, analysis for 3-months or even1-year intervals being more relevant.
Apart from these improvements,there is also a substantial reduction ofthe computational time for LM model(ca. 20 minutes for 24 h forecast) incomparison with HRM model (about50 minutes for 24h forecast), for thedomain and resolution considered in thisstudy.
TESTING OF THE NON-HYDROSTATIC MODEL LM ON THE ROMANIAN TERRITORY25
At fine resolution, the modelaccurately reproduces the local sub-diurnal cycle of 2m-temperature and therotation of wind, but the values of forecastsurface fields present systematicdifferences when compared with theanalyzed period observations, suggesting
again the necessity of fine tuning on theconsidered domain.
Acknowledgements: The authors are grateful tothe two anonymous reviewers for theirconstructive remarks.
a) b)
c) d)
Fig. 11. Wind vector (arrows) at 2.8 km resolution for 03.11.2004, anticipation of: a) 6h; b) 12h; c) 18h;d) 24, superimposed on domain orography, contour interval 100m.
L. VELEA, I.V. PESCARU, R. DUMITRACHE, I.K.MOUSTAFA
26
REFERENCES
Burridge, D. M. (1975)A split semi-implicit reformulation of the Bushby-Timpson 10-Level model.Quart. J. Roy. Meteor. Soc. 101, 430, 777-792.
Davies, H. C. (1976)A lateral boundary formulation for multi-level prediction models. Quart. J. Roy.Meteor. Soc. 102, 432, 405-418.
Domms, G. (2001)A schme for monotonic numerical diffusion in the LM. Cosmo Technical Report,No.3,Deutscher Wetterdienst, Offenbach, Germany
Domms, G . and SCHATTLER, U. (2001)The Multi-Layer Version of the DWD Soil Model TERRA_LM. Technical ReportNo. 2, Deutscher Wetterdienst, Offenbach, Germany
Domms, G . and Schattler, U. (2002)A Description of the Nonhydrostatic Regional Model LM. Part I: Dynamics andNumerics, Deustcher Wetterdienst, Offenbach, Germany, 134 pp.
Domms, G., Forstner, J., Heise, E., Herzog, H. J., Raschendorfer, M., Schrodin, R.,Reinhardt, T. and Vogel, G. (2004)
A Description of the Nonhydrostatic Regional Model LM. Part II: PhysicalParameterization, Deustcher Wetterdienst, Offenbach, Germany, 120 pp.
Jacobsen, I. and Heise, E. (1982)A new economic method for the computation of the surface temperature innumerical models. Contr. Atmos. Phys., 55, 128-14.1
Kain, J.S. and Fritsch, J.M. (1993)Convective Parameterization for Mesoscale Models: The Kain-Fritsch Scheme. In:The Representation of Cumulus Convection in Numerical Models. MeteorologicalMonograph No. 46, American Meteorological Society, 165-170.
Louis, J.-F. (1979)A parametric model of vertical eddy fluxes in the atmosphere. Bound. LayerMeteor., 17, 187-202
Majewski, D. (2005)HRM-User’s Guide, Deustcher Wetterdienst, Offenbach, Germany, 98 pp
Mellor, G.L. and Yamada, T. (1974)A hierarchy of turbulence closure models for planetary boundary layers. J. Atmos.Sci., 31, 1791-1806
Ritter, B and Geleyn, J.F. (1992)A comprehensive radiation scheme for numerical weather prediction models withpotential application in climate simulations. Mon. Wea. Rev., 120 , 303-325.
Schrodin, R., Edelmann, W., Majewski, D., Schaetller, U., Doms, G., Ritter, B., Link,A., Hanisch, T. (1995)
Documentation of the EM/DM System. Deutscher Wetterdienst, Offenbach,Germany
Skamarocl, W.C. and Klemp, J.B. (1992)The stability of time-split numerical methods for the hydrostatic and thenonhydrostatic elastic equations. Mon Wea. Rev., 120, 2109-2127
TESTING OF THE NON-HYDROSTATIC MODEL LM (LOKAL MODELL) ON THE ROMANIANTERRITORY
27
Stanski, H.R., Wilson, L.J. and Borrows, W.R. (1989)Survey of common verification methods in meteorology, WWW, Technical Report,No. 8, WMO/TD., No. 358.
Tiedtke, M., (1989)A comprehensive mass flux scheme for cumulus parameterization in large scalemodels. Mon. Wea Rev., 117 , 1779-1799.
Thomas, S., Girard, C, Doms, G. and Schattler, U. (2000)Semi-implicit scheme for the DWD Lokal-Modell, Meteor. Atmos. Phys.,75, 105-125
Wicker, L. and Skamarock, W. (1998)A time-splitting scheme for the elastic equations incorporating second-orderRunge-Kutta time differencing. Mon. Wea Rev., 126, 1992-1999
