precipitation in the northern alpine region: case-study

Deutsches Zentrum fuÈr Luft- und Raumfahrt (DLR), Institut f�ur Physik der Atmosph�are, Oberpfaffenhofen, Germany

Precipitation in the Northern Alpine Region: Case-Study-TypeValidation of an Operational Forecast Model

C. Keil and H. Volkert

With 10 Figures

Received March 2, 1999Revised July 28, 1999

Summary

Quality and limitations of current quantitative precipitationsimulations are determined using the Deutschland-Modellof Deutscher Wetterdienst (DWD). The model independentvalidation data comprise the regular rain-gauge network ofDWD, taylored composites of an international radarnetwork, Doppler winds from the DLR research radar,and meteorological timeseries at a ground station. Areaskill scores are compared for convectively dominated aswell as more synoptically forced situations in the northernAlpine region. Inspection of the temporal evolution of thecomponents comprising the atmospheric water budget andof the precipitation partition gives insight in the differentnature of excitation mechanisms of heavy precipitation. Thesimulations are found to exhibit growing potential forfollow-on hydrological applications, while a real break-through appears to necessitate the coming generation ofnon-hydrostatic operational forecast models with increasedspatial resolution.

1. Introduction

The Alpine precipitation climatology exhibitscharacteristic seasonal variations in the northernAlpine region, where enhanced precipitation isobserved during the main convective periodfrom May until September (Frei and Sch�ar,1998). Distinct seldom occuring events of heavyprecipitation, like the development of mesoscaleconvective systems or rainfall episodes enhancedby orographic blocking, contribute signi®cantlyto the annual mean precipitation and are able to

trigger natural disasters as, e.g., wide area¯ooding, ¯ash ¯oods, landslides or serious casesof hail damage. Suf®ciently precise forecasts ofsuch severe precipitation events one to three daysin advance could signi®cantly contribute to avoidloss of life and reduce damage.

Presently regional numerical weather predic-tion (NWP) models are operationally applied togain short-range forecasts with a horizontalspatial resolution approaching 10 km. With thecurrently available computer resources, suchhigh resolution models are only feasible for alimited domain, i.e. in the regional or meso-scale. Thus the large-scale meteorological evolu-tion has to be passed on through time dependentlateral boundary conditions. The atmosphericprocesses which control precipitation to a largeextent act on the meso-scale (typical lateralextents between 1 and 1000 km) and render aprecise rainfall prediction considerably moredif®cult than, e.g., a pressure, wind or tempera-ture forecast. Many complex atmosphericprocesses resulting in a highly variable spatio-temporal precipitation distribution can onlypartly be explicitly incorporated in such meso-scale models. The remaining unresolved processeshave to be parameterized and pose the cumulusparameterization problem (see, e.g., the overviewarticle by Molinari and Dudek, 1992). Conse-quently precipitation shows the greatest

Meteorol. Atmos. Phys. 72, 161±173 (2000)

sensitivity to the model physics and its predictioncontinues to be one of the great challenges inatmospheric modelling.

A detailed assessment of an atmosphericmodel's capabilities constitutes one importantquality control step. The validation of theprecipitation distribution, in particular, representsa substantial task because of its aforementionedintrinsic spatial and temporal variability. More-over evaluation dif®culties emerge from thesevere gap between the spatial density oftraditionally used rain gauge observations andthe calculated data. Radar data, on the otherhand, offer precipitation estimates with highspatial resolution and have not yet been widelyused for NWP model validation. Up to veryrecently the routine veri®cation of precipitationby the national weather services was restricted toseasonal rainfall means validated with precipita-tion observations at selected synoptic stations.Such a method veils information about themodel's accuracy for individual episodes ofsevere precipitation. The latest, more preciseevaluations undertaken by European meteorolog-ical services are not available in the openliterature. Therefore it is felt that our assessmentregarding a few episodes possesses signi®cantexplorative character.

Reliable forecasts of the mesoscale precipita-tion patterns are seen as an essential prerequisitefor physically based hydrological forecastingschemes including (¯ash) ¯ood warning proce-dures with a lead time of two or more days,particularly in mountainous terrain. Within theproject HERA regional simulations of fournorthern Alpine episodes with considerablerainfall amounts were performed with the opera-tional forecast model of DWD (Table 1). One ofthem is discussed here in detail to highlight thepotential of quantitative precipitation forecasts aswell as its actual limitations.

After a short description of the forecast modelan overview of the synoptic scale weathersituations is presented. Subsequently the simula-tion results are compared with various back-ground observations comprising high resolutionrain gauge measurements, taylored composites ofa precipitation radar network and timeseries ofmeteorological parameters recorded at a groundstation. The radar re¯ectivities are augmented byDoppler wind velocities to provide a spatial highresolution data set for an evaluation of thesimulated ¯ow ®eld. The results of this con-vective case are juxtaposed to other northernAlpine precipitation episodes whose rainfallpatterns were induced by prominent weathersituations. Finally the model is used as adiagnostic tool to investigate the atmosphericwater budget in a pre-Alpine river catchment.The main ®ndings and a brief outlook aresummarized in the concluding section.

2. Model Description and Initial Conditions

The numerical simulations are performed withthe hydrostatic Deutschland-Modell (DM; Schro-din, 1997). Introduced in 1993, this regionalmodel is the main short range weather forecast-ing tool applied by DWD till the end of 1999.The model domain covers central Europe (Fig. 1)encompassing 109� 109 grid points with ahorizontal mesh size of �x� 14 km. Verticallythe model's atmosphere is discretized in 20layers on a hybrid coordinate system.

Physical processes associated with cumulusconvection are divided in resolvable-scale pro-cesses and subgrid-scale processes. Grid-scaleprecipitation includes parameterized cloudmicrophysics of the Kessler type and allows forinteractions between water vapour, cloud water,rain and ice. Subgrid-scale deep convectionprocesses are parameterized by the traditional

Table 1. List of Investigated HERA Cases. The Precipitation Amounts (RRobs) Denote the Observed Daily Area Average and theObserved Maxima (RRmax) with their Locations (Latitude �max, Longitude �max)

Date Synoptic Type RRobs

(mm/d)RRmax

(mm/d)�max

(deg.)�max

(deg.)

5 July 1996 cold front with intense prefrontal squall 14.3 89.0 48.2 7.98 July 1996 cyclogenesis over the Alps 24.2 124.9 47.7 12.34 July 1994 Mesoscale Convective System 10.8 96.5 47.8 10.114 July 1994 Mesoscale Convective System 6.4 57.9 49.1 8.2

162 C. Keil and H. Volkert

Tiedtke mass ¯ux scheme, designed originallyfor a coarse resolution, where moisture conver-gence in the sub-cloud layer is the essentialmechanism to generate convection (Tiedtke,1989).

Six-hourly analyses taylored for DM wereused as initial and boundary conditions. A Daviestype boundary relaxation scheme is applied toadjust the model variables towards the large scalelateral boundary conditions. The model runswere initialized at 00 UT for every individualcase. Data of the period � 6 h till � 30 h of thesimulation are considered, i.e. only values afterthe model spin-up time.

We note that we cannot determine the entirepredictability as if we had used truely forecastboundary conditions, as e.g., in Mladek et al.(this volume). On the other hand, all discrepan-cies found here have to be attributed to modelde®ciencies, e.g., insuf®cient horizontal resolu-tion or too simple physical parameterisations, orto an inappropriate mesoscale analysis scheme.

3. Synoptic Overview

On 5 July 1996 a mesoscale cyclone developedover France embedded in a strong upper tropo-spheric southwesterly air¯ow and graduallymoved across Germany while still intensifying.Ahead of the cold front warm air was prevailingin southeastern Germany which favoured the

formation of deep convection. Along a prefrontalconvergence line strong thunderstorm activitywas observed.

Subsequently a surface low developed over theAlpine region under the leading edge of an upperlevel trough extending to the western Mediterra-nean on 7 July 1996 00 UT. At the border ofwarm air masses over eastern Central Europe andsigni®cantly cooler ones to the west a widespread precipitation region formed. This area ofintense rainfall moved northeastward across theCzech Republic and Poland on 8 July 1996 alongthe still quoted `̀ route Vb'' of last century'scatalogue of European storm tracks (Bebber,1891). Additionally the precipitation got intensi-®ed by orographic lifting processes in thenorthern Alpine region. Under the in¯ux of coolbut humid maritime air the temperature insouthern Germany only slightly exceeded 10 �Con 8 July 1996.

In contrast to those pronounced meso-�-scaleweather situations two mesoscale convectivesystems (MCS) were observed on 4 July and 14July 1994 during the Severe ThunderstormExperiment (SETEX). Emerging from scatteredthunderstorms above the Jura and Black Forestmountains rapidly growing MCSs evolved duringprefrontal conditions and moved eastward towardsBavaria. Their precipitation signatures had morelocal character resulting in lower rainfall arealaverages (Table 1).

4. Observations and Simulationsfor 5 July 1996

The simulation quality of DM for a distinctprecipitation event is now exempli®ed in detailfor the rainfall episode on 5 July 96. Thepresentation of observational data follows adecrease in space and time scale and comprisesdaily precipitation totals in southern Germany,the overall temporal development and motion ofthe precipitating system as seen by an interna-tional radar composite, Doppler-derived radialwinds viewed from a single radar station, andlocal timeseries.

4.1 Daily Precipitation Totals

Traditionally rain gauge observations are used toevaluate the temporally integrated rainfall over

Fig. 1. Domain of DM with coastlines, political bound-aries, a 2� 2� geographical grid, and topographic heights ofthe model orography (grey scale)

Precipitation in the Northern Alpine Region: Validation of an Operational Forecast Model 163

24 hours. We ®rst note the different nature of thedata types. Each model value at a horizontalresolution of 14 km represents an average over anarea of about 200 km2, whereas the actualprecipitation can vary signi®cantly over such anarea, especially above complex terrain andduring convectively dominated conditions. Onthe other hand the discrete character of raingauge observations limits their representativitywhich is in¯uenced by, e.g., the highly variablecharacter of convective precipitation as well aslimitations due to the exact location of the gaugewithin the topography. Nonetheless rain gaugeobservations are widely used for a quantitativeprecipitation evaluation.

In Fig. 2 the precipitation distribution on5 July 1996 is depicted. First of all the spatiallyheterogeneous character of the convective eventbecomes evident. The strongest precipitation wasobserved in the Black Forest mountains insouthwestern Germany with a total rainfall of89 mm. The maximal simulated precipitationyields 56 mm just two mesh widths further west.Another zone with considerable precipitation liesnortheastward of Munich where deep convectionalong the convergence line resulted in rainfalltotals of more than 30 mm. In the modelprediction a comparable region of enhancedprecipitation appears 50 km further to the eastwith maxima amounting to about 20 mm.

Averaging the irregularly spaced 1031 avail-able rain gauge observations onto the model gridallows a more quantitative comparison. Gener-ally the model slightly overestimates the arealrainfall by 9.6% in southern Germany on 5 July1996 (see Table 2). Statistical measures of

precipitation forecast accuracy are provided inthe next section.

4.2 Temporal Overall Development

Now, we examine the spatial and temporalstructure of the precipitation ®elds. Observationsof the weather radar networks of the Alpinecountries, which are assembled to the HERANorth Alpine Radar Composite (Hagen et al.,this volume), are qualitatively compared withsimulated pseudo-radar images. Once again it isnecessary to keep the different nature of the twodata types in mind.

Radar detectability of hydrometeors can beseriously affected by their distances from theradar site and by orography intersecting the radarpulses. The main error sources of radar derivedrain rates are due to beam blocking, beambroadening and the decrease of re¯ectivity withheight (Joss et al., 1995). In the HERA NorthAlpine Radar Composite the radar precipitationestimates are assigned the maximum precipita-tion intensity found in a given vertical column,no matter whether it was detected by one orseveral radars of the network. Thus it is notpossible to determine the height of the observedhydrometeors. The precipitation estimates aregrouped into six logarithmically scaled intensitylevels on a 2 km horizontal grid.

The DM pseudo-radar images display max-imum precipitation ¯uxes sampled from thevertical columns of the model grid. The depictedinstantaneous precipitation ¯ux results fromprecipitation forming processes within onemodel time step. The model precipitation rates

Fig. 2. Observed (left) versus simulated (right) 24 h precipitation totals for 5 July 1996


are discretized according to the radar intensityclasses for this investigation. In contrast tothe higher spatial resolution of the radar datathe simulated precipitation ¯ux is only availableon the model grid with a `̀ pixel size'' of14� 14 km2.

In Fig. 3 the radar observed convective activityis compared with its simulated counterpartdescribed above. The radars detected a pro-nounced frontal line with imbedded convectioncores, which moved eastward across southernGermany. The leading edge of the convectiveline approached Munich at 17 UT with highre¯ectivity values of 55 dBZ indicating heavyprecipitation or even hail.

A similiar though patchier structure of aneastward progressing frontal line with highprecipitation rates is reproduced by the model.The simulated precipitating region lags theobservation by about one hour and crosses theLake of Constance area at 17 UT. Precipitatingprocesses along that convergence line yieldrainrates of up to 31 mm/h. The upward motion,one requirement for the precipitation formation,reaches values up to 1.1 m/s along that frontalline. Relatively warm and humid air ahead ofthe front sustains the formation of convection.Mesoscale ascent of the humid air yieldsprecipitation rates of 16 mm/h at 19 UT whenthe convergence line has already crossedMunich. The heights, in which the precipitation¯ux attains its maximum, vary during this periodbetween 770 hPa and 885 hPa. Below that layersubcloud evaporation reduces the quantity ofcondensate reaching the earth's surface. With thefrontal passage the wind is veering towards westand cooler and drier air gets advected.

The qualitative comparison of radar compo-sites and numerically simulated pseudo-radarimages reveals the model's ability to capture the

general structure and progression of the conver-gence line.

Aligned mesoscale convective areas withupward motion are simulated; these are favour-able for the initiation and maintainance ofconvection. Moreover the instantaneous precipi-tation ¯ux illustrates that Tiedtke's convectiveparameterization scheme applied in the meso-scale produces a rather patchy precipitationstructure. These rainrates result from the con-vergence of moisture within one single modeltime step. Since the convective cloud watercontent is a diagnostic variable, no life cycle ofconvective clouds is included and, thus, thesnapshots of the depicted precipitation ¯ux showan incoherent structure. Integrating the instanta-neous rainrates over time yields, e.g., hourlyrainfall accumulations with smoother structures(cf. Volkert, this volume; Fig. 7). But at agridsize of 14 km a snapshot of the precipitation¯ux at a speci®c time step of the modelintegration cannot be expected to show as manydetails of the convective activity as detected bythe radars.

4.3 Doppler Winds

Data taken by DLR's Dopplerized polarizationdiversity radar (POLDIRAD) are used to retrievethe radial wind ®eld relative to the radar locationin Oberpfaffenhofen. Doppler wind measure-ments are possible when precipitation particlesare present in the air volume under examination.Therefore no velocity data are available in the`dry' (grey coloured) region ahead of the front(Fig. 4). Peak radial velocities were detected in azone 20 to 50 km WSW (255�) of the radar site,where the Nyquist-corrected signal amounted to24 m/s in an altitude of about 400 m aboveground. The region in which the ¯ow is

Table 2. DM-Simulated Precipitation of fourHERACases comprising Area Averages RRDM, Relative Contribution of ConvectiveRainfall RRconv, Ratio of Modelled Versus observed Precipitation RRDM/RRobs and the modelled Maxima with their Locations(Latitude �max, Longitude �max)

Date RRDM

(mm/d)RRconv

(%)RRDM/RRobs

(%)RRmax

(mm/d)�max

(deg.)�max

(deg.)

5 July 1996 15.63 91 109.6 55.5 48.4 7.88 July 1996 19.94 24 82.2 123.5 47.5 10.04 July 1994 7.49 94 69.0 76.6 47.8 10.214 July 1994 12.55 85 196.2 133.3 48.7 11.1


perpendicular with respect to the radar site, i.e.the radial velocity diminishes, extends NNWwith an angle of 330�. The southern branch ofthat zone is masked due to the absence ofhydrometeors. Ahead of the convergence lineweak northerly winds of 2 m/s prevailed south ofthe radar.

In Fig. 5 the radial component of the simulatedhorizontal wind ®eld relative to the radar site isshown for a layer 480 m above ground. Mainradar detected features of the ¯ow ®eld can beidenti®ed in the model data. The maximum of thesimulated radial wind velocity is situated WSW(240�) of the radar site, where the southwesterly

Fig. 3. Radar composites of the Swiss, Austrian and German Radar network on 5 July 1996 (left) and instantaneous pseudoradar images of the Deutschland Modell simulation (right; colour coded as the radar composites)


¯ow is directed straight towards the virtualobserving station and peaks in 21 m/s. The zonewith zero radial velocity values extends fromNWN (295�) to SSW (190�) with respect to thevirtual radar site. A slight northerly componentof the wind vector is predicted for an area about30 km south of Munich and compares well withthe measurements. The secondary maximum atthe southern boundary of the depicted region iscaused by the Alpine orography, where the windvelocity 480 m above ground exceeds 16.3 m/s.

To the authors' knowledge comparisonsbetween radar Doppler winds and meso-scale¯ow simulations were sofar not documentedfor Central Europe. The reasonable agreementbetween Figs. 4 and 5 is felt as an encouragingone as reliable motion ®elds are a prerequisite forgood precipitation forecasts.

4.4 Local Timeseries

Timeseries of meteorological parameters offeranother model independent data source tovalidate the simulation and to quantify possibletiming or phase errors which are frequentlycontained in the initial data of the simulation.

The weather station at DLR is located atop thebuilding about 15 m above ground. Its timeseriesclearly mark the frontal passage (Fig. 6). At 1700UT the pressure reached its minimum followedby a sharp increase of 4 hPa within 15 min. Thetemperature concurrently dropped by 8 K downto 16 �C. The gust front crossed the observationplatform with wind velocities up to 16 m/s nineminutes later (1709 UT) followed by a narrowzone of heavy precipitation which resulted in11 mm within 6 min.

Comparing the observed timeseries (data every3 min.) with the simulated ones (data every12 min.) at the closest gridpoint the muchsmoother character of the simulation becomesobvious. The initial and ®nal values of tempera-ture, pressure and wind speed of the depicted 12hourly timeseries agree remarkably well. Acloser look reveals a time difference of one hourbetween observation and simulation. At 18 UTall of the displayed meteorological parametersindicate the passage of the frontal line. Simulta-neously with the steady temperature drop,the pressure and the wind speed increase.The accumulated precipitation caused by the

Fig. 4. Doppler radial wind velocity measured by POL-DIRAD at Oberpfaffenhofen on 5 July 1996, 16:49 UT with1� elevation. Negative values indicate a ¯ow directedtowards the radar. For technical reasons values below theNyquist-threshold (ÿ 13.6 m/s) are wrapped to the otherside of the colour scale (as due W of the radar site)

Fig. 5. Simulated radial wind velocity relative to the grid-point Oberpfaffenhofen (colour coded as the Doppler radialwind in Fig. 4) for 5 July 1996, 18:00 UT and 480 m aboveground. The arrows indicate the Cartesian ¯ow components(max. vector equals 25 m/s). A time-lag of 1 h betweenobservations and simulation is taken into account (see sub-section 4.4)


convergence line amounts to less than 2 mm inthe model.

Clearly the local scale structure and strength ofthe frontal discontinuity cannot be completelysimulated at a grid size of 14 km. But again thecoincidence between locally observed and grid-point time-series is considered as an importanthint for the feasibility of reliable meso-scalemodelling of severe weather events.

5. Results for other HERA Cases

After the close-up looks at details of the weatherevolution of 5 July 1996, we now apply statisticalmeasures of precipitation totals to investigate themodel's capabilities more generally and presentresults of other HERA rainfall episodes.

First the precipitation totals of 8 July 1996 arepresented in Fig. 7. The uniformity of the valuesis in some contrast to the much more spatiallystructured convective pattern three days earlier.The observed, quasi-uniform precipitation dis-tribution had maxima at the northern Alpine rimwhich amounted to 124.9 mm in 24 hours (seeTable 1). The predicted maxima of 123.5 mmcompares well in amplitude, although it isdisplaced by 150 km to the west in the northernAlpine foreland (see Table 2). Orographic liftingprocesses triggered by the Alps in the southward¯ow cause these extreme rainfall totals.

The predicted distribution of rainfall totals lessthan 20 mm along the Upper Danube valley andprecipitation accumulations of more than 40 mmonly several kilometers northwestward in theSwabian Jura are con®rmed by rain-gaugeobservations. Between 49� and 50� latitude theriver Rhine bounds the rainfall region to the westand agrees well with available observationaldata. Only in eastern Bavaria, in an area betweenrivers Danube and Inn, the model predicts lessthan the observed precipitation totals. Hence theDM underpredicts the area averaged rainfall insouthern Germany by 18%.

Unfortunately there is no single measure offorecast skill that completely characterizes theaccuracy of a precipitation prediction. To quan-tify the accuracy of the precipitation totals weuse the bias score and the threat score of aspeci®ed precipitation threshold amount at agiven grid point during a speci®ed time period(Anthes, 1983). The bias score measures thetendency of a model to forecast too small or toolarge an area of a given amount of precipitation,while the threat score describes the skill insimulating the exact area of precipitation totalsabove a given threshold. These areas agreecompletely if the scores amount to one. A biasscore greater than one indicates an overpredic-tion of precipitation.

The calculated scores in Fig. 8 illustrate thecase-to-case variability. Generally the bias is

Fig. 6. Observed (bullets) and simulated (diamonds) time-series of temperature, pressure difference (relative to anarbitrary offset), wind speed and precipitation (from top tobottom) in Oberpfaffenhofen on 5 July 1996. In the timeinterval 16 to 19 UT the abscissa is stretched


increasing for greater precipitation amounts. Biasscore values for rainfall thresholds greater than40 mm may not be meaningful since the numberof points receiving this amount is often small(e.g., only 2 grid cells on 5 July 1996, bottomgraph in Fig. 8). The bias score shows anoverestimation of the precipitation totals exceed-ing 20 mm for 5 July 96 whereas a slightunderrating can be identi®ed for low rainfallthresholds. For the stratiform rainfall event of 8July 1996 the bias score shows good agreementfor precipitation accumulations up to 40 mm.Only scores greater than this threshold reveal themodel's underestimation. The threat scores showa decreasing skill with increasing threshold aswell. For a threshold of 10 mm the threat scoresrange from 0.85 to 0.17 for 8 July 1996 and 14July 1994, respectively. For the 20 mm thresholdthey range from 0.55 to 0.025. As the areadiminishes in size with increasing precipitationthreshold correct forecasts become more dif®cultto obtain. Consequently best scores are deter-mined for the episode on 8 July 1996 withwidespread precipitation in southern Germany,when 50 grid points received more than 40 mm.Moreover a quite good threat score is evident forthe case of 5 July 1996, discussed in detail above,where the score is close to one for precipitationtotals of 2 mm and still 0.5 for the 10 mmthreshold.

On the contrary, the cases of July 1994 areexhibiting low scores and this points to the notsuf®ciently accurate representation of the pre-cipitation processes. These episodes were ofmore local character (see, e.g., the number ofgrid cells receiving 1 mm in Fig. 8) and were

Fig. 7. Observed (left) versus simulated (right) 24 h precipitation totals for 8 July 1996. The Isar-Amper catchment as referred toin Figs. 9 and 10 is surrounded by the bold rectangle

Fig. 8. Rainfall scores for 4 HERA cases versus a prescribed24 h precipitation threshold. The bias score, threat score andthe number of grid cells (top to bottom) are depicted for5 July 1996 (squares), 8 July 1996 (bullets), 4 July 1994(diamonds) and 14 July 1994 (triangles). Note the compres-sion of the ordinate for high bias score values (BS > 2). Thenumber of cells receiving more than the threshold amount isbased on observations averaged onto the model grid and isscaled logarithmically


induced by rapidly growing mesoscale convec-tive systems. Although DM exhibits some skill incapturing the meso-scale evolution of the ¯ow®eld (for details see Keil and Volkert, 1999) themodel fails in reproducing the correct precipita-tion pattern. Sensitivity experiments (not dis-played) show the substantial in¯uence of thetime of model initialisation and particularly thecrucial character of the humidity ®elds on theprecipitation totals.

Hence, regarding scores of precipitation totals,DM is better suited for the simulation of preci-pitation events induced by pronounced largermeso-scale ¯ow structures (July 1996 cases) thanfor events developing on the smaller meso-scale.The large variations found re¯ect the complexityof the precipitation generation processes and thedif®culty of quantitatively predicting rainfallpatterns.

6. Diagnostic Study for a Water Shed

After the examination of the model's ability topredict mesoscale precipitation ®elds in southernGermany we proceed using the DM as adiagnostic tool to inspect the atmospheric waterbudget. The reasonably good agreement withobservations quali®es the precipitation episode inearly July 1996 to exemplify the analysis ofmoisture ¯ux. Before the examination of theatmospheric water budget we ®rst investigate thepartition of convective and grid-scale precipita-tion in some more detail.

The considered domain comprises the catch-ments of the rivers Isar and Amper south of theircon¯uence near Freising in southern Bavaria. Itcovers 45 model grid-meshes with an area ofabout 9000 km2 in the northern Alpine region(see Fig. 7). In Fig. 9 two striking precipitationevents are evident within the 4 day episode from5 July 1996, 06 UT until 9 July 96, 06 UT. Acloser inspection reveals their different nature.The convective precipitation dominates in theevening hours on 5 July and is correlated withstrong area averaged upward motion up to23 cm/s prevailing over about 50% of thesubdomain (data are taken from the 680 hPapressure level in a height of about 3 km). Incontrast the second rainfall extremum on 8 Julyis of grid scale nature. Steady upward motion of

less than 10 cm/s over the whole subdomainleads to heavy rainfall. This partitioning resultsin 91% convective precipitation on 5 July versusonly 24% on 8 July (Table 2).

While the precipitation event on 5 July isidenti®ed with the passage of the strongconvergence line, the second maximum on 8July is caused by quasi-stationary orographicinduced lifting processes as mentioned in theprevious section. Thus the model captures thedifferent excitation mechanisms of heavy pre-cipitation during this 4 day episode.

Following an approach for climate studies(Sch�ar et al., 1999), we now inspect the atmo-spheric water budget which comprises thefollowing terms:

FqD � Fqw � EV ��Qÿ RR � Res

where Fqi represents the netto ¯ux of moisture�Fqi � Fqi;in ÿ Fqi;out; qD: water vapor; qw:liquid water), EV evapotranspiration, �Q thestorage of water in the control volume, RRprecipitation, while Res designates a residualwhich ideally should vanish. The verticallyintegrated moisture ¯ux is averaged over thealready mentioned subdomain of the Isar-Ampercatchment.

The temporal evolution of the moisture ¯uxshows a rather complex and spiky behaviourduring the passage of the convergence line on 5July (Fig. 10). This can partly be attributed to thediscrete nature of the budget calculation fromhourly model output ®elds and partly to the smallsize of the subdomain comprising only 5� 9 gridcells. Nonetheless some insight can be gainedfrom the displayed timeseries of the main termsforming the atmospheric water budget. While the®rst spell of rain at 14 UT is dominated by themoisture advection FqD the second precipitationevent at about 20 UT on 5 July mainly originatesfrom the lifting of the humid air (stored in �Q)which was advected (FqD) earlier that day. Asshown in Fig. 9 the strong upward motion periodis very well correlated with episodes of heavyprecipitation. Unless strong upward motion ispresent in the small control volume the moistureadvection FqD is mainly balanced by the storageterm �Q (e.g., on 7 July). The orographicprecipitation on 8 July is nourished by acombination of moisture advection and therelease of stored humidity. The advection of


cloud water FqW is negligible throughout thewhole episode (not displayed). Autochthonousin¯uences, through term EV, contribute compara-tively little to overall RR partly owing to thesmallness of the control volume. However the

dependence of the evapotranspiration on thesolar insolation with its maximum around noonbecomes clearly evident. On 8 July evapotrans-piration is reduced by a compact cloud systemwhich inhibited insolation.

Fig. 10. Timeseries of the moisture ¯uxes contributing to the atmospheric water budget averaged over the Isar-Ampercatchment for the same 4 day episode as in Fig. 9; explanation of letter codes in the text

Fig. 9. Timeseries of the convective (RRk) and grid scale (RRs) components of precipitation and the vertical velocity (w, in680 hPa or about 3 km; only upward motion) together with its areal exteant [A(w)] in the Isar-Amper catchment from 5 July1996, 06 UT until 9 July 1996, 06 UT


7. Conclusions

The thorough case-study-type inspection ofsimulated precipitation signatures gives indica-tions of the model's ability to properly representthe circulation features that actually occurred innature. Observational data from different sourcescomprising traditionally used rain gauge mea-surements, timeseries of meteorological param-eters and radar data were complementary usedfor the assessment of DM's capabilities tosimulate different precipitation regimes.

Using these various observations a detailedmodel validation for the 5 July 1996 case studywas performed. An inspection of the 24 hourlyprecipitation distribution shows minor differ-ences and the error in the mean accumulatedrainfall amounts to less than 10%. The horizontalprogression of the precipitation ®elds, as com-pared with radar re¯ectivity data, shows themodel's ability to catch the fundamental frontalcharacteristics in the meso-scale. Yet applyingTiedtke's convective mass ¯ux scheme in themeso-scale yields rather patchy instantaneousrain-rate signatures. The identi®ed temporaldiscrepancy of about one hour is not infrequentin mesoscale modelling and explained by de®cien-cies in the data at the model boundaries (Haase et al.,1997; Majewski, 1997) and the insuf®cient modelresolution. Particularly the comparison of radarderived wind data and the simulated winds showsthe capability to reproduce the observed meso-scale¯ow ®eld reasonably well. On the other hand thetimeseries elucidate the limitation of the hydrostaticmodel in precisely describing the sudden passage ofa well developed convergence line.

Subsequently the bias and the threat scoreswere applied to obtain compact quantitativemeasures of the model's quality to capturedifferent precipitation regimes. These exactingscores illustrate the larger skill of the model tosimulate precipitation structures of meso-�-scaleextent, i.e. higher scores were obtained for the 5July and 8 July 1996 than for the two MesoscaleConvective Systems in 1994 (Fig. 8).

The inspection of the temporal evolution of thepartition of precipitation in combination with themean upward vertical velocity averaged over thecontrol domain in the northern Alpine forelandrevealed the model's skill in capturing thedifferent excitation mechanisms of precipitation

during the 4 day episode in early July 96.Diagnosing the components of the atmosphericwater budget gave further insight in theirdifferent behavior during varying synoptic con-ditions.

Finally the Deutschland Modell was found tobe well suited for the simulation of precipitationevents induced by pronounced meso-scale ¯owstructures as was illustrated by the reported caseof early July 1996 and the initiation phase of theOder ¯oods the following year (see Keil et al.,1999). The model's limitations are reached whensimulating the MCSs of 1994. Here the DMfailed to predict the timing of the precipitation aswell as the rainfall amount in the northern Alpineregion. Sensitivity studies outline the importanceof the initialisation time and its humiditydistribution.

In summary, we conclude that the quantitativesimulation of precipitation has growing potentialto provide valuable information for a follow-onhydrological simulation for suf®ciently largeriver basins and aiming at ¯ood warnings. Thevalidation techniques, which were presented here,should be further applied to monitor the atmo-spheric simulation quality. An important break-through is envisaged when non-hydrostatic simu-lation models with solely explicit precipitatingice-phase processes on horizontal resolutions of afew kilometres, for which research applicationsare available (Stein et al., this volume; Benoitet al., 2000), will be transferred to operational use.

Acknowledgements

We thank Detlev Majewski (DWD, Offenbach) for hissupport with the Deutschland Modell and for providing theDM-analysis data. Daniel L�uthi (ETH, Z�urich) kindly madeavailable the code for the budget calculations. Two refereesprovided valuable comments to strengthen the line ofargumentation. This study was in parts supported by theEuropean Commission, Programme `̀ Environment andClimate'' under contract ENV4-CT96-0332.

References

Anthes, R. A., 1983: Regional models of the atmosphere inmiddle latitudes. Mon. Wea. Rev., 111, 1306±1335.

Bebber, W. J. van, 1891: Die Zugbahnen der barometrischenMinima. Meteorol. Z., 8, 361±366 (plus xii plates).

Benoit, R., Pellerin, P., Kouwen, N., Ritchie, H., Donaldson,N., Joe, P., Soulis, R., 2000: Toward the use of coupledatmospheric and hydrologic models at regional scale.Mon. Wea. Rev., (accepted).


Frei, C., Sch�ar, C., 1998: A precipitation climatology of theAlps from high-resolution rain-gauge observations. Int. J.Climatol., 18, 873±900.

Joss, J., Lee, R., 1995: The application of radar-gaugecomparisons to operational precipitation pro®le correc-tions. J. Appl. Meteor., 34, 2612±2630.

Haase-Straub, S. P., Hagen, M., Hauf, T., Heimann, D.,Peristeri, M., Smith, R. K., 1997: The squall-line of 21July 1992 in southern Germany: An observational casestudy. Beitr. Phys. Atmos., 70, 147±165.

Hagen, M., Schiesser, H.-H., Dorninger, M., 2000: Monitor-ing of mesoscale precipitation systems in the Alps and thenorthern Alpine foreland by radar and rain gauges.Meteorol. Atmos. Phys., 72, 87±100.

Keil, C., Volkert, H., Majewski, D., 1999: The Oder ¯ood inJuly 1997: Transport routes of precipitable water diag-nosed with an operational forecast model. Geophys. Res.Lett., 26, 235±238.

Keil, C., Volkert, H., 1999: Radar composites and mesoscaleprecipitation forecasts: Determination of model de®cien-cies through case studies. Proceed. Final Conf. of COST75 activity: Advanced weather radar systems, CEC, Lux-embourg, ISBN 92-828-4907-4, 407±417.

Majewski, D., 1997: Operational regional prediction.Meteorol. Atmos. Phys., 63, 89±104.

Molinari, J., Dudek, M., 1992: Parameterization of convec-tive precipitation in mesoscale numerical models: Acritical review. Mon. Wea. Rev., 120, 326±344.

Sch�ar, C., L�uthi, D., Beyerle, U., Heise, E., 1999: The soil-precipitation feedback: A process study with a regionalclimate model. J. Climate, 12, 722±741.

Schrodin, R. (ed.), 1997: Quarterly Report of the operationalNWP-ModelsoftheDeutscherWetterdienst.AvailablefromDWD, Postfach 100465, D-63004 Offenbach, Germany.

Stein, J., Richard, E., Lafore, J. P., Pinty, J. P., Ascencio, N.,Cosma, S., 1999: High-resolution non-hydrostatic simula-tions of ¯ash-¯ood episodes with grid-nesting and ice-phaseparameterization. Meteorol. Atmos. Phys., 72, 203±221.

Volkert, H., 2000: Heavy Precipitation in the Alpine Region(HERA): Areal rainfall determination for ¯ood warningsthrough in-situ measurements, remote sensing and atmo-spheric modelling. Meteorol. Atmos. Phys., 72, 73±85.

Tiedtke, M., 1989: A comprehensive mass ¯ux scheme forcumulus parameterization in large-scale models. Mon.Wea. Rev., 117, 1779±1800.

Authors' address: Christian Keil and Hans Volkert,DLR, Institut f�ur Physik der Atmosph�are, D-82234Wessling, Germany (e-mail: [email protected],[email protected]).


precipitation in the northern alpine region: case-study

Documents