Studies on Data Uncertainties of the Local Model of the German Meteorological Service in the Catchment Area of the River Ilm

Home , Ilm (Thuringia)

Advances in Geosciences, 5, 113–118, 2005 SRef-ID: 1680-7359/adgeo/2005-5-113 Advances in European Geosciences Union Geosciences © 2005 Author(s). This work is licensed under a Creative Commons License.

Studies on data uncertainties of the local model of the German Meteorological Service in the catchment area of the river Ilm

K. Pfannschmidt1,2 and H. Dohler¨ 2 1Thuringer¨ Landesanstalt fur¨ Umwelt und Geologie, Prussingstraße¨ 25, 07745 Jena, Germany 2Friedrich Schiller University of Jena, Ernst-Abbe-Platz 2, 07743 Jena, Germany Received: 7 January 2005 – Revised: 1 August 2005 – Accepted: 1 September 2005 – Published: 16 December 2005

Abstract. To assess the prognostic values concerning pre- ventive flood protection in Thuringia. Strong efforts were cipitation and air temperature in the river catchment area of made to improve the protection of the population and to the Ilm/Thuringia issued by the German Meteorological Ser- avoid flood damages, for example by modernising the flood vice, the differences between the prognostic values and re- warning and signalling service (Hack, 2003). In the Free lated measured values were statistically evaluated. The in- State of Thuringia, the monitoring network on the water- vestigations covered the period from November 1999 to De- courses comprises 176 gauges, 52 of which having been se- cember 2003. The spatial resolution corresponds to the LM lected and established as flood gauging and signalling sta- grid cells and the available data. The observed precipitation tions (Spanknebel et al., 1999). values were corrected with regard to systematic wind error in The technical developments of the last years implemented dependence of temperature. in the German Meteorological Service have led to an im- The forecast data were investigated comprehensively in or- provement of the climate forecast data. The numerical der to quantify the statistical reliability. Therefore, the 2nd weather prediction model (Local Model, LM) enables the moment as the case may be the standard deviation of the dif- model-based short-term discharge prediction (URL[3]). This ferences between observed precipitation values and forecast- constitutes the basis for the implementation of a flood warning precipitation values have been calculated. The investi- ing and information service for a prototypical Thuringian gations were subdivided into the complete stretch of time, flood gauging and signalling station. The LM forecast data the hydrological summer and the hydrological winter. Each shall form the basis of a model-based operational discharge measured value was analysed with regard to the 12-, 24-, 36- prediction with the aim to detect looming flash-floods in and 48-hour forecast made by the DWD. good time and to be able to inform accordingly. As a result, empirical analytical functions were developed For this purpose, the gauge of Grafinau-Angstedt¨ located to describe adequately the statistical quantities for standard in the upper reaches of the Ilm river has been selected be- deviation of precipitation and temperature for the periods of cause of the experiences and practical results already ex- hydrological summer and hydrological winter in dependence isting for this partial catchment area (TEG) at the chair on forecasting time, mean ground level and roughness (frac- of Geoinformatics of the Institute for Geography of the tal dimension) of the LM-grid cell. Friedrich Schiller University of Jena (Staudenrausch, 2000; Fritzsche, 2001) For hydrological modelling of the TEG of Grafinau-Angstedt¨ the Precipitation-Runoff Modelling Sys- tem/Modular Modelling System (PRMS/MMS) (Leavesley 1 Introduction and Stannard, 1995) will be used based on the distributive approach of the Hydrological Response Units (HRUs). After the devastating flood disaster of August 2002 in Ger- The prognostic uncertainty calculated for the hourly pre- many, the Czech Republic, and Austria, many activities in cipitation and temperature forecast values will be used as σ- the field of flood protection were launched. borders in the operational working of a prognostic discharge This event showed clearly that early warning systems are model for the TEG of Grafinau-Angstedt.¨ the most effective instrument to prevent disasters due to the generation of a valuable gain in time for far-sighted ac- tion (URL[1]). New impetus was also given to the pre-

Correspondence to: K. Pfannschmidt ([email protected]) 114 K. Pfannschmidt and H. Dohler:¨ Studies on data uncertainties

• Fig. 1. Ground level andFig. selected 1 stations Ground ( ) with level catchment and boundaryselected of thestations Ilm (–). (•) with catchment boundary of the Ilm (-)

2 The local model of the German Meteorological Ser- tal climatic influences on the other hand (Thuringian Basin) vice (DWD) The mean annual temperatures(Bongartz, are 2001). around 5° C in the Thuringian Forest, 7° C in the Ilm- The mean annual temperatures are around 5◦C in the The LM is an operational,Saale-Ohrdruf hydrodynamical weatherzone forecastand 8° ThuringianC in the Forest, Inner-Thuringian 7◦C in the Ilm-Saale-Ohrdruf lowland. zone andThese differences are caused model solving a prognostic equation for the vertical move- 8◦C in the Inner-Thuringian lowland. These differences are 2 ment. A square grid composedby the ofdifferent elementary areas altitudes of 49 km andcaused windward/lee by the different effects altitudes and with windward/lee the prevailing effects south-west winds in the covers the whole arealow of Central mountain Europe. (URL[2]) range. The With Lo- regardwith the prevailingto precipitation, south-west winds the in study the low mountainregion has to be subdivided into a cal Model forecast started in November 1999. The model is range. With regard to precipitation, the study region has to be able to resolve weather-relevantlow mountain local landscape and details a summer and subdivided rain into type. a low mountainIn the and cool a summer and rain damp type. In zones of the ridge of the small-scale meteorological structures. Additionally to the the cool and damp zones of the ridge of the Thuringian For- horizontal and verticalThuringian wind components, Forest, pressure andprecipitation soil est, precipitation of up of to up to1200 1200 mm/year mm/year is reached, is on reached, the lee on the lee side of the moisture, the LM provides air temperature values and hourly side of the Thuringian Forest in the direction of the central liquid and solid precipitationThuringian totals. In Forest this connection in the 48- directionThuringian of Basin the the central annual precipitation Thuringian is below 500Basin mm the annual precipitation is hour model calculationsbelow are carried 500 out mm that provide (Fig. current 2). The(Fig. mean 2). The meanannual annual snow snow cover cover depth isdepth around 400–is around 400-500 mm in the forecast data, refreshed every 12 h. These forecast data shall 500 mm in the Thuringian Forest – a region with abundant form the basis of a model-basedThuringian operative Forest discharge - a predic- regionsnowfall with andabundant mostly with snowfall settled snow conditionsand mostly – around with settled snow conditions tion with the aim to detect looming flash-floods and to pro- 100–300 mm in the mountain foothills and <100 mm in the vide adequate information-, around for early warning100-300 systems. mm in theThuringian mountain Basin – foothills an area with littleand snowfall<100 (Bongartz,mm in the Thuringian Basin - an area with little snowfall. [Bong01]2001). 3 The river catchment area of the Ilm

As prototypical river catchment area (FEG) the Ilm river 4 Analysed parameters and investigation period catchment in Thuringia, in particular the partial catchment area (TEG) of Grafinau-Angstedt¨ located in the upper The prognostic LM data were analysed with regard to their reaches, has been selected. statistical quality for the catchment area of the Ilm river in With a flowing length of about 135 km, the Ilm is a rela- Thuringia. The main focus was put on precipitation and air tively small river draining an area of about 1025 km2. temperature as main input parameters for hydrological mod- The difference in altitude between the source and the els. The investigations covered the period from the beginning mouth is 848 m (see Fig. 1). The mean annual discharge of the LM forecast in November 1999 to December 2003. of the Ilm is 6.59 m3/s. From the climatic point of view, For the study area, a representative selection – concerning Thuringia is located within the temperate zone of transition, the spatial distribution – of LM grid cells to be analysed was characterized by the maritime coastal climate with Atlantic determined, taking into consideration the measurement sta- influences on the one hand (Thuringian Forest) and continen- tions available.

Fig. 2 Mean annual precipitation in the Ilm catchment area with significant LM-grid cells

Analysed Parameters and Investigation Period The prognostic LM data were analysed with regard to their statistical quality for the catchment area of the Ilm river in Thuringia. The main focus was put on precipitation and air temperature as main input parameters for hydrological models. The investigations covered the period from the beginning of the LM forecast in November 1999 to December 2003. For the

Fig. 1 Ground level and selected stations (•) with catchment boundary of the Ilm (-)

The mean annual temperatures are around 5° C in the Thuringian Forest, 7° C in the Ilm- Saale-Ohrdruf zone and 8° C in the Inner-Thuringian lowland. These differences are caused by the different altitudes and windward/lee effects with the prevailing south-west winds in the low mountain range. With regard to precipitation, the study region has to be subdivided into a low mountain and a summer rain type. In the cool and damp zones of the ridge of the Thuringian Forest, precipitation of up to 1200 mm/year is reached, on the lee side of the Thuringian Forest in the direction of the central Thuringian Basin the annual precipitation is below 500 mm (Fig. 2). The mean annual snow cover depth is around 400-500 mm in the Thuringian Forest - a region with abundant snowfall and mostly with settled snow conditions -, around 100-300 mm in the mountain foothills and <100 mm in the Thuringian Basin - an study area, a representative selection - concerning the spatial distribution - of LM grid cells to K. Pfannschmidt andbearea H. analysed D withohler:¨ Studieslittlewas determined, snowfall. on data uncertainties [Bong01]taking into consideration the measurement stations available. 115

Statistical Investigations The deviations of the forecast data from the measured data were investigated. The forecast was subdivided into four 12-hour periods in accordance with the generation rhythm of the LM forecast. Concerning the seasons, the investigation period was subdivided into the complete stretch of time, the hydrological summer and the hydrological winter. Regarding the aspect of temporal resolution, the measured data available allowed a rating of deviations in daily and hourly sums for precipitations, and in hourly values for temperatures. In addition to the analysis of the distribution functions of differences between observed and forecasting values, the statistical parameters of expected value µ, 2nd moment and σ-borders were determined (Fig. 3). Altogether, 28 measurement stations were included in the investigation, of which 4

stations (Dornburg, Großobringen, Schmücke, Oberweißbach) are providing the measured Fig. 2 Mean annual precipitation in the Ilm catchment area with significant LM-grid cells Fig. 2. Mean annual precipitationdata for precipitation in the Ilm catchment and temperature area with signiﬁcant as hourly LM-grid values. cells.

Fig. 3. Schematic overview of the statistical investigations carried out. Fig. 3 Schematic overview of the statistical investigations carried out

5 Statistical investigations Oberweißbach) are providing the measured data for precipi- To facilitate comparability, the hourly precipitationtation and temperature forecasts as hourlyfor rain values. and snow available for the period from the beginning of the LM forecastTo facilitate in November comparability, 1999 the hourlyto December precipitation 2003, fore- were The deviations of the forecast data from the measured data casts for rain and snow available for the period from the be- were investigated.aggregated The forecast to was daily subdivided sums of into precipitation four values. In this connection, the CET-based time of ginning of the LM forecast in November 1999 to Decem- 12-hour periods inreading accordance the with measured the generation values rhythm of the day and the UTC-based time of the forecast had to be ber 2003, were aggregated to daily sums of precipitation val- of the LM forecast. Concerning the seasons, the investiga- considered. Optimised correction functionsues. In thiswere connection, determined the CET-based to correct time the of readingwind theerror tion period was subdivided into the complete stretch of time, measured values of the day and the UTC-based time of the the hydrological summeraccording and theto hydrologicalRichter. [Richt95] winter. Re- The temperature values required were calculated from the forecast had to be considered. Optimised correction func- garding the aspect ofvalues temporal obtained resolution, from the measuredthe measurement data stations according to the method of the arithmetic tions were determined to correct the wind error according to available allowed a rating of deviations in daily and hourly distance weighted mean. After that, theRichter temperature (1995). The values temperature were valuesregionalised required wereapplying calcu- the sums for precipitations, and in hourly values for tempera- saturated adiabatic temperature coefficient.lated from the values obtained from the measurement stations tures. In addition to the analysis of the distribution func- according to the method of the arithmetic distance weighted tions of differencesApplying between observedsuitable and methods, forecasting the val- corrected hourly and daily precipitation sums were mean. After that, the temperature values were regionalised ues, the statistical parameters of expected value µ, 2nd mo- regionalised with regard to the calculatedapplying average the saturatedlong-term adiabatic precipitation temperature mean coefﬁcient. of the related ment and σ-borders were determined (Fig. 3). Altogether, 28 measurement stations were included in the investigation, 4 of which 4 stations (Dornburg, Großobringen, Schmucke,¨ grid cell for the period from 1970 to 2000. For that purpose the quotient of LM-cell mean and the precipitation mean of the station was evaluated. Now, the differences between the daily LM sums and the corrected and regionalised readings were analysed. Therefore, empirical distribution functions, the expected value and the 2nd

moment M2 were determined. The 2nd moment is a quadratic distance measure with regard to the value zero. Using the basic

parameter M2 ensures the absolute comparability of the stations with each other . σ = Fig. 4 indicates the root-mean-square deviations σ ( M 2 ) for different stations (hydrological summer and winter separated). For comparison, the altitude profile of the 116respective stations is shown in the background. K. Pfannschmidt and H. Dohler:¨ Studies on data uncertainties

Fig. 4. Root-mean-squareFig. deviation4 Root-mean-square of the daily precipitation deviation of [mm/day] the daily values precipitation with altitude [mm/day] proﬁle values of the stationswith altitude subject profile to investigation. of the stations subject to investigation

Fig. 5 shows the distribution of the root-mean-squarethe absolute deviations comparability for ofthe the complete stations withperiod each in other.the river catchment area subject to investigation.Figure √ 4 indicates the root-mean-square deviations σ (σ= M2) for different stations (hydrological summer and winter are separated). For comparison, the altitude proﬁle of the respective stations is shown in the background. Figure 5 shows the distribution of the root-mean-square deviations for the complete period in the river catchment area subject to investigation. The statistical quality calculated this way was assessed with regard to special regional features in view of position and altitude as well as seasonal differences. Concerning the distribution functions the conclusion was drawn that an optimum adaptation to the analysed distributions was feasible by means of the Lorentz function in Eq. (1) (Fig. 6). Fig.Fig. 5 5. Spatial Spatial distribution distribution of the of root-mean-square the root-mean-square deviations deviations (rms) (rms) of the daily precipitation values in the area 5 a of the dailysubject precipitation to investigation values in the area subject to investigation. f (x) = (1) + − 2 1 b(x µ) The statistical quality calculated this way was assessedwith with the expected regard value to specialµ and the regional half-value features width . Applying suitable methods, the corrected hourly and daily As the temporal resolution of the prognosticated discharge inprecipitation view of position sums were and regionalised altitude withas well regard as to seasonal the cal- differences.of the forecast model must be equal to the temporal resolu- culated average long-term precipitation mean of the related Concerning the distribution functions the conclusiontion was of thedrawn input LMthat forecast an optimum data, it was adaptation also necessary to grid cell for the period from 1970 to 2000. For that purpose consider the statistical quality of the hourly forecast data. tothe the quotient analysed of LM-cell distributions mean and was the precipitation feasible by mean means of of theThe Lorentz hourly precipitation function valuesin Eq. taken (1) into(Fig. consideration 6) the station was evaluated. were regionalised to the respective LM grid, such as the daily Now, the differences between the daily LM sums and the sums. The subsequent statistical quality rating proceeded in corrected and regionalised readings were analysed. There-a analogy to the daily sums. fore, empirical distribution functions, thef ( expectedx ) = value and To rate the statistical quality of the temperature(1) forecasts, 1+ b(x − µ) 2 the 2nd moment M2 were determined. they were analysed with an hourly resolution. Again, the dif- The 2nd moment is a quadratic distance measure with references between LM and the measured values were analysed gard to the value zero. Using the basic parameter M2 ensures in analogy to the evaluation of precipitation. with the expected value µ and the half-value width 1 . b

As the temporal resolution of the prognosticated discharge of the forecast model must be equal to the temporal resolution of the input LM forecast data, it was also necessary to consider the statistical quality of the hourly forecast data. The hourly precipitation values taken into consideration were regionalised to the respective LM grid, such as the daily sums. The subsequent statistical quality rating proceeded in analogy to the daily sums. To rate the statistical quality of the temperature forecasts, they were analysed with an hourly resolution. Again, the differences between LM and the measured values were analysed in analogy to the evaluation of precipitation. Resulting from that, propositions on the quality of the LM forecast data as function of time and location of stations in the hourly and daily time resolutions could be made. The analysed distribution functions of the hourly values show a stronger correlation with the Lorentz function than those of the daily values (Fig. 7).

K. Pfannschmidt and H. Dohler:¨ Studies on data uncertaintiesFig. 6 Analyzed distribution function of station Dornburg 117 for daily values

Fig.7 Analyzed distribution function of station Dornburg for hourly values Fig. 6 Analyzed distribution function of station Dornburg for daily Fig. values 7. Analyzed distribution function of station Dornburg for Fig. 6. Analyzed distribution function of station Dornburg for daily hourly values. values. Each of these investigations was made for the 4 forecast periods of the first 12 hours, the Resulting from that, propositions on the quality of the LM second,rj =f dj − third,2 and fourth 12 hours. forecast data as function of time and location of stations in Ashj –there the altitude was to of thebe LMexpected grid cell a dependence of the forecast quality upon the roughness of the the hourly and daily time resolutions could be made. The hmittel – is the mean value of the altitude of all LM grid cells analysed distribution functions of the hourly values show a surfaceanalysed relief and the ground level, the fractal dimension D (Eq. 2) was calculated for all stronger correlation with the Lorentz function than those of significantn – empirical LM exponent grid cells [Barn88]: the daily values (Fig. 7). 1 Each of these investigations was made for the 4 forecast hj n H n = d ln[N(l)] with the property (3)= D , (2) periods of the first 12 h, the second, third, and fourth 12 h. Dj= − h M l d mittelln[l] As there was to be expected a dependence of the forecast quality upon the roughness of the surface relief and the withPrecipitation N(l) being – hydrological calculated summer: according to the box-counting algorithm [Falc90]. groundFig.7 level, Analyzed the fractal distribution dimension functionD (Eq. of station 2) was Dornburg calculated for hourly values = + 3+ σSo(τ, j) 0, 41 rj 0, 429 Hj for all significant LM grid cells (Barnsley, 1988): (4) τ −0, 04 r + 0, 061 566 H 3 d ln[N(l)] j j DEach= − of these investigationswith the property wasM =madelD for the (2)4 forecastResults periods of the first 12 hours, the d ln[l] Precipitation – hydrological winter: second, third, and fourth 12 hours. In the course of evaluating the results, analytical functions were developed to describe with N(l) being calculated according to the box-counting al- = + 1 + As there was to be expected a dependence of the forecastsufficientlyσWi(τ, quality j) 0 the,upon327 statisticalrj the0 ,roughness3873 quantitiesHj of thein dependence on the forecast period, the ground level gorithm (Falconer, 1990). + 1 1.8 − 1 surface relief and the ground level, the fractal dimensionand the D surface(Eq.τ 0 ,2)011 profile. wasrj calculated3 (0, 518 Hj for all0, 3873 Hj ) 6significant Results LM grid cells [Barn88]: The formulae given below for the root-mean-square(5) deviation of the dayly values were developed with values being equalised through 30 stations and in case of hourly values In the course of evaluating the results, analytical functions Temperature – hydrological summer: d ln[N(l)] D D = − with the property M = l , through 4 stations. (2) were developed to describe sufficiently the statistical quan- = + −7 + d ln[l] σSo(τ, j) 0, 421 rj 1,n8 Hj tities in dependence on the forecast period, the ground level The basis functions H and rj was estimated by non-linear optimisation of the parameters hj j −7 (6) and the surface profile. τ 0, 01 rj + 0, 1066 H with N(l) being calculated according to the box-countingand algorithm fdj (see below) [Falc90]. relative to thej calculated σ - values. The empirical parameters in Eqs. (4- The formulae given below for the root-mean-square deviation of the dayly values were developed with values being 7)Temperature result from – hydrological linear winter:fit to the chosen basis functions. For smart description of the equalised through 30 stations and in case of hourly values = + −7+ σWi(τ, j) 0, 428 rj 1, 804 Hj throughResults 4 stations. n −4 1 −5 −7 The basis functions H and r was estimated by non-linear τ 6.66 · 10 rj + (1, 982 H − 1, 804 H ) In the course of evaluatingj j the results, analytical functions were developed3 to describej j 7 optimisation of the parameters hj and f dj (see below) rel- (7) ativesufficiently to the calculated the statisticalσ -values. quantities The empirical in dependence parameters on the forecast period, the ground level inand Eqs. the (4–7) surface result profile. from linear fit to the chosen basis functions.The Forformulae smart description given below of the uncertaintiesfor the root-mean-squareσ in Eqs. (4– 7 deviation Summary of the dayly values were 7) auxiliary variables was introduced. Equation (3) describes thedeveloped generalised with dependence values ofbeing the altitude equalised related through to a mean 30 stationsIn the majority and ofin thecase analysed of hourly stations, values the σ value of daily value.through 4 stations. precipitation in the hydrological summer exceeds that of the The following definitionsn apply: hydrological winter (except for some mountain stations of The basis functions H j and rj was estimated by non-linear optimisation of the parameters hj j – index of the LM grid cell σ the Thuringian Forest). tand[h] –fd forecastingj (see below) time τrelative=(t/12− to1) the calculated - values. TheTheσ empiricalvalues of precipitation parameters in in the Eqs. summer (4- increase only f7) d j result– the fractal from dimension linear offit the to LM the grid chosen cell, basis functions.insignificantly For withsmart the description altitude (n=3, seeof Eqs.the 3 and 4) , in

7 118 K. Pfannschmidt and H. Dohler:¨ Studies on data uncertainties the winter there is a stronger dependence on the altitude (n=1 Falconer, K.: Fractal Geometry, John Wiley & Sons, Chichester, respectively n=1.8 in Eq. 5). New York, Brisbane, Toronto, Singapore, 38–42, 1990. The σ values of temperature are inversely proportional to Fritzsche, C.: Vergleichende ereignisbezogene Modellierung der the altitude. Here, the altitude values have only an insignifi- Abflussbildung mit PRMS/MMS und TOPMODEL in zwei cant influence in summer and winter (n=5 respectively n=7, kleinen Quelleinzugsgebieten im Thuringer¨ Wald, Diplomarbeit, see Eqs. 6 and 7). In all precipitation and temperature val- FSU Jena, 2001. Hack,H.-P.: Vorbeugender Hochwasserschutz in Thuringen,¨ ues analysed, the influence of the surface roughness is of the Thuringer¨ Ministerium fur¨ Landwirtschaft, Naturschutz und same scale. The dependence of forecast time is nearly linear. Umwelt, 2003. The distribution functions of the differences between mea- Leavesley, G. H. and Stannard, L. G.: The Precipitation-Runoff sured values and forecasting values follow the Lorentz line. Modelling System – PRMS, in: Computer Models of Watershed This property is even more striking in the hourly values. In Hydrology, edited by: Singh, V. P., Water Resources Publica- general, hourly values reveal greater deviations in summer, tions, 281–310, 1995. increasing with the altitude/precipitation. The influence of Richter, D.: Ergebnisse methodischer Untersuchungen zur microclimatic conditions – such as the prevailing wind di- Korrektur des systematischen Messfehlers des Hellmann- rection, local warming, etc. – has not been included in the Niederschlagsmessers, Berichte des Deutschen Wetterdienstes present studies due to the lack of available data. 194, Selbstverlag des DWD, Offenbach am Main, 1995. It is possible to calculate the uncertainties of precipi- Spanknebel, G., Kaufmann, M., Kowalski, B., and Kelm, R.: Mod- tation and temperature by LM-forecasting for each grid ernisierung des Hochwasserwarn- und Meldedienstes im Freis- cell with the developed analytical functions. On this taat Thuringen,¨ Wasserwirtschaft/Wassertechnik, 4, pp. 47, 1999. way the total uncertainty of the hydrological prognostic Staudenrausch, H.: Untersuchungen zur hydrologischen Topologie discharge model can evaluate from the uncertainty of the von Landschaftsobjekten fur¨ die distributive Flussgebietsmodel- model itself and the calculated uncertainties of the input data. lierung, Dissertation, FSU Jena, 2000. URL[1]: Deutsches Komitee fur¨ Katastrophenvorsorge e.V.: http: Edited by: P. Krause, K. Bongartz, and W.-A. Flugel¨ //www.dkkv.org/, last access: 09.11.2004. Reviewed by: anonymous referees URL[2]: Deutscher Wetterdienst DWD: Numerische Wetter- vorhersagemodelle des DWD-Das Lokalmodell LM, http:// www.dwd.de/de/FundE/Analyse/Modellierung/lm.htm, last ac- References cess: 09.11.2004. URL[3]: Buchholz, O.: Hochwasservorhersage auf Basis eines Barnsley, M.: Fractals everywhere, Academic Press, London Inc., Echtzeitbetriebs von NASIM, Hydrotec-Ingenieurgesellschaft 172–182, 1988. fur¨ Wasser und Umwelt, http://www2.hydrotec.de/, last access: Bongartz, K.: Untersuchung unterschiedlicher 09.11.2004. Flachendiskretisierungs-¨ und Modellierungskonzepte fur¨ die hydrologische Modellierung am Beispiel Thuringer¨ Vorfluter, Dissertation, FSU Jena, 2001.