Remote Sensing nd Geogrphic Informtion Systems for Design nd Opertion of Wter Resources 1 57 Systems (Proceedings of Rbt Symposium S3, April 1997). IAHS Publ. no. 242, 1997 GISrelted bseflow simultion for wter blnce nd precipittionrunoff modelling in the River Rhine bsin PETER KRAHE, KARLHEINZ DAAMEN, RAINER MULDERS & KLAUS WILKE Federl Institute of Hydrology, PO Box 309, D56003 Koblenz, Germny Abstrct For the ppliction of wter blnce nd precipittionrunoff models in lrge river bsins, like the Rhine bsin, regionliztion of model prmeters using GISfcilities is necessry. For the Moselle, which is the min tributry of the Rhine, distributed prmeter estimtion technique for the recession constnt bsed on multivrite regression between physiogrphic dringe bsin chrcteristics nd conventionl nlysis of recession curves is tested. The use of the regionlized recession constnt in monthly wter blnce model bsed on the wellknown ThornthwiteMther method is shown. It cn be concluded tht despite of the uncertinty of the proposed method due to the use of smllscle hydrogeologicl nd geologicl mps the "regionlized recession constnt" cn be used for bseflow simultion in wter blnce nd precipittionrunoff models. The method described is lso helpful tool for testing the relibility of dily dischrge dt nd the results of hydrogrph nlyses in lrge river bsins. INTRODUCTION The extreme floods in 1993/1994 nd 1995 s well s some dry yers in the lst decde hve cused lrge economicl dmge in the highly urbnized nd industrilized Rhine bsin. To mitigte this dmge one of the strtegies tht is under discussion is the improvement of existing wrning nd forecsting systems. Severl studies hve been initited by the public wter uthorities of the riprin countries long the Rhine to provide necessry hydrologicl foundtions. Of specil interest re nthropogenic flood cuses nd the impct of possible climte chnge on the dischrge of the Rhine nd its min tributries. Some of the themes mentioned bove re elborted under the uspices of the Interntionl Commission for the Hydrology of the River Rhine (CHR/KHR). It ws decided to build common GIS dtbse covering the whole Rhine bsin, especilly for the ppliction of the wter blnce models in climte impct studies nd precipittionrunoff models to supplement existing forecsting systems. A generl problem besides the selection of the hydrologicl model is the estimtion of the model prmeters in lrge river bsins. Different prmeter estimtion techniques re presently in use tking GIS fcilities into ccount. As n exmple for distributed prmeter estimtion the use of multivrite regression technique using bsin chrcteristics nd results of conventionl hydrogrph nlysis is tested. Therefore, the recession constnt s key prmeter for simulting bseflow in wter blnce nd precipittionrunoff models is selected.
158 Peter Krhe et l. It will be shown tht the recession constnt cn be linked to simple monthly wter blnce model which is presently in use for the Rhine bsin to crry out severl detiled studies concerning the selection of sptil computtion units or interpoltion methods of hydrometeorologicl vribles. GIS AND HYDROMETEOROLOGICAL DATABASE For the interntionl Rhine bsin (185 000 km 2 ) shown in Fig. 1, digitl dtbse of physiogrphic nd hydrometeorologicl dt, the RHINEGIS, ws built up (Tble 1) in collbortion with the public wter uthorities of the riprin countries Switzerlnd, Frnce, Germny, Luxembourg nd The Netherlnds. The projected scle of the digitl mps within the RHINEGIS is 1:500 000 for vector dt nd for grid (rster) dt. As shown in Tble 1, this scle could not yet be relized for ll dt becuse of lck of nlogue mps. The high costs for some dtsets, especilly for the Digitl Elevtion Model nd for lnduse dt with higher resolution, re further obstcle. On the other hnd, for the Moselle bsin with n re of 28 152 km 2 dtbse with higher resolution nd scles rnging from 1:20 000 to 1:50 000 nd 50 m 50 m grid is in preprtion. This dtbse cn be used to crry out severl sensitivity nlyses concerning e.g. the selection of the required sptil resolution with respect to predefined model pplictions.
GISrelted bse/low simultion/or wter blnce nd precipittionrunoff modelling 159 Tble 1 GIS dtbse of the River Rhine bsin. Mp ttribute Digitl Elevtion Model Lnduse clssifiction in 13 clsses Digitl Elevtion Model nd derived géomorphologie prmeters Soil mp nd soil physicl properties Hydrogeologicl mp Hydrogeologicl units Bsin res River network Dt type Point Grid Vector 30 m* Loction nd storge cpcities of lrge reservoirs x Loction nd properties of guging nd meteorologicl sttions x Longterm men monthly climtologicl mps x * Only vilble for the Moselle. 1:1000 000 1:1500 000 1:1 500 000 1:500 000 1:200 000* 1:50 000* s.. nd 1:1000 000 The soil lyer (Commission of the Europen Communities, 1985) nd the hydrogeologicl lyer (CHR/KHR, 1978) s well s subbsin boundries nd the river network hve been digitized from nlogue mps by different ntionl nd interntionl uthorities. The subbsin boundries re clssified with n hierrchicl system which llows for ggregtion of subbsins to lrger computtion units using GIS fcilities. Digitl Elevtion Models were prepred by the public uthorities of the riprin countries nd compiled for storing in the GIS. For ll countries except Switzerlnd lnduse dt re derived by clssifiction of LndstTM dt. In regrd of the high costs for the stellite derived lnduse clssifiction for the Germn, French, nd Luxembourg prts of the Rhine bsin the clssified high resolution lnduse dtset ws ggregted on grid using filter technique (Krhe & Mendel, 1993). The selected lnduse clsses re urbnized, densely urbnized, lkes, grsslnd, griculturl lnd, coniferous forest, deciduous forest, mixed forest, perennil plnting, snds or dunes, moorlnd nd heth, specilly used res, nd rocks or boulders. All dt re stored in the RHINEGIS using ARC/INFO, while in this study the SPATIAL ANALYSIS SYSTEM (SPANS) softwre is employed. The hydrometeorologicl dtbse is still in preprtion nd includes meteorologicl nd climtologicl dt such s monthly men, monthly, dily, nd 3 to 6 hourly vlues of precipittion, temperture, reltive humidity, wind speed, nd durtion of sunshine s well s output from the wether forecst model "DEUTSCHLANDMODELL" of the Germn Wether Service (DWD). A system for storing nd retrieving meteorologicl dt bsed on reltionl dtbse mngement system is in preprtion. Dily nd hourly dischrge dt re lso collected. For the Germn prt of the Rhine bsin, dily dischrge dt of bout 500 guging sttions for the time period from 1961 to 1990 hve been prepred. For the interntionl Moselle bsin discussed in this study dtset of 80 guging sttions with dily dischrge dt for the time period 19711980 hs been processed.
160 Peter Krhe et l. WATER BALANCE MODEL The wter blnce model used in the Rhine bsin (Krhe et l., 1996) is vrition of scheme tht ws first proposed by Thornthwite & Mther (1957) nd subsequently trnsposed in tody's nottion with slight modifiction by Willmot et l. (1985) nd Vôrôsmrty & Moore (1991). Runoff cn only occur if precipittion (P) exceeds the potentil évpotrnspirtion (PE) plus the moisture deficit between ctul moisture content nd wter holding cpcity (WHO). In this cse ctul évpotrnspirtion ( ) equls PE nd the surplus wter fills the runoff storge (SR). The computtion of the monthly runoff (/?, ) in the originl scheme follows the eqution: *,. =KR i. l +(lk)sr i (1) where i = number of the month, K = runoff coefficient. The production of both direct nd delyed runoff is not seprtely considered in the previous schemes. The model ssumptions do not llow for direct runoff during summer months. In the Rhine bsin, except for the higher mountinous regions, this restriction leds to n overestimtion of E. First pplictions of the model in the Moselle bsin (Krhe et l., 1996) hve shown this effect nd did lso indicte tht urbn res produce certin mount of direct flow resulting in n underestimtion of runoff, especilly during the summer months. Further improvements of the originl scheme were obtined by: () Introducing direct runoff, produced by urbn res, by removing frction IM of the rinfll incident t the surfce (IM = frction of urbn re to sptil computtion unit). (b) Considering infiltrtion excess runoff, occurring if precipittion intensity exceeds the locl infiltrtion cpcity. This process is dominnt especilly in the cse of convective rinfll during summer months. Additionl to IM the prmeter FR reflecting this process is introduced to remove portion of monthly rinfll mount. Presently this prmeter is determined by clibrtion. Considering () nd (b), 7?, of eqution (1) cn be interpreted s bseflow nd the runoff coefficient K s recession constnt. Interflow is not tken into ccount t this time. DISTRIBUTED PARAMETER ESTIMATION Different pproches of prmeter estimtion techniques considering soclled distributed or semidistributed wter blnce nd precipittionrunoff models re in use. A common strtegy is to link model prmeters nd physiogrphic chrcteristics of the bsin. Therefore, the ppliction of GIS becomes centrl prt in prmeter estimtion. Tody's efforts in developing physicllybsed hydrologicl models re to define model prmeters directly from bsin chrcteristics. Opposite to this, the present prctice is relte model prmeters to physiogrphic chrcteristics nd clibrte the whole set of prmeters by optimiztion techniques. This procedure often leds to
GISrelted bse/low simultion/or wter blnce nd precipittionrunoff modelling 161 unrelistic vlues of the prmeters. To model lrge river systems like the River Rhine, prmeter estimtion for unguged res is required. Some prmeters like WHC or impervious res cn be clculted directly by GIS ppliction. Other prmeters hve to be regionlized by regression nlysis. Therefore estimted prmeters nd physiogrphic chrcteristics re considered in guged subbsins. The size nd number of subbsins depend on the sptil nd temporl vribility of the hydrologicl processes. The clculted empiricl reltionships re used to obtin sptilly distributed prmeters nd cn be pplied to the whole bsin, i.e. lso in unguged res. In the following, this methodology is tested for deriving the recession constnt s key prmeter for simulting bseflow in wter blnce nd precipittionrunoff models. Estimtion of recession constnt In the originl scheme of the wter blnce model, Thornthwite & Mther (1957) used constnt vlue of 0.5 for K. Interpreting i?, s bseflow, K is recession constnt vlid for monthly dischrge dt tht cn be estimted by K = e om where k is the recession constnt obtined by nlysing recession curves of dily dischrge dt (Q) using the Boussinesq eqution Q = Ôo *" (3) with n = number of dy nd Q 0 s dischrge t n = 0. Demuth (1993) developed two methods DEREC1 nd DEREC2 for estimtion of the recession constnt k bsed on this procedure. Furthermore, Demuth exmined severl regression equtions between k nd bsin chrcteristics for regionliztion of this prmeter. In one of these equtions the res of hydrogeologicl units (HG1...HG14) re used s independent vribles k = b 0 HG1"HG2" 2 HG3" 3...HG14"' 4 (4) where bo... bl4 re regression coefficients. Demuth (1993) found sesonl dependency of the prmeter k, resulting in one set of regression coefficients for the summer (April to September, k s ) nd nother for the winter (October to Mrch, k w ). This method is wellsuited for GIS ppliction. The 14 HGs re determined from n overly of hydrogeologicl mp (lyer) contining informtion on the cpcity of groundwter extrction s well s depths of the quifer nd geologicl mp (lyer). For exmple, the dominnt hydrogeologicl units in the Moselle bsin re HG5 nd HG13. HG5 mens tht the cpcity of groundwter extrction is less thn 1000 m 3 dily extrctble wter nd the depth of the quifer is deep. The geologicl formtion consists of the Trissic series Muschelklk nd Bunter Sndstone. For HG13 the cpcity of groundwter extrction per dy is less thn 100 m 3 nd the geologicl formtion is crystlline bsement. The depths of groundwter quifers cn be shllow s well s deep. (2)
162 Peter Krhe et l. 8 0T 50 25' / «_.1 f ' /^g::<7;.;:p:~>' î 75 km 47045 5 20' Fig. 2 Mp of hydrogeologicl units in the River Moselle bsin. <f^~j /.v j. ia<p«/i...,.. r " > J? 4 : I ivdrogeologicl 1 iiits Lke ]HG2 3HG3 ]HG4 LV.Î HG 5 IDHG6 3HG8. 3HG12 QHG13 subbsins Appliction The study re is the River Moselle bsin. A lrge portion of the bsin re uplnds rnging from 300 m up to 1400 m.s.l. in the Vosges. The bsin is chrcterized by the low wter holding cpcity of the soils s well s low groundwter storge cpcity. This my led to low wter periods of long durtion s well s extreme floods in winter months. A mp of 14 hydrogeologicl units ws constructed following the rules described by Demuth (1993) using vilble dtsets from the RHINEGIS. Becuse of the bsence of digitl geologicl mp the following procedure hs been pplied: () Reclssifiction of the digitl hydrogeologicl mp to 14 clsses representing the groundwter cpcity nd depth of quifer using the GIS fcilities. (b) Overlying plot of the new mp over n nlogue geologicl mp with scle
GISrelted bseflow simultion for wter blnce nd precipittionrunoff modelling 163 o ks x kw U.3/ 0.96 0.95 o 0.94 CO CO 0.93 D) 0) ^ 0.92 I > 0.91 o D x xx x RI x * 9 x x no ŒDTCDO Ï&C^ x «0.90 0.89 0.88 0.87 ~i 1 1 0.88 0.89 0.90 0.91 0.92 0.93 0.94 0.95 0.96 0.97 k by DEREC2 Fig. 3 Sctterplot of recession constnts estimted by regression vs. hydrogrph nlysis. 1:2 150 000 (CHR/KHR, 1978) by hnd nd reclssifiction of the mp interctively within the GIS. (c) The res of hydrogeologicl units within selected subbsins were determined by GIS overly nlysis. The resulting mp is shown in Fig. 2. Given the res of HG1 to HG14 within subbsin, the recession constnts were estimted using eqution (4). On the other hnd, method DEREC2 s outlined by Demuth (1993) ws used for clculting the dily recession constnt by hydrogrph nlysis. The prmeter k ws determined by fitting eqution (3) to the mster recession curve which ws constructed from observed bseflow recessions. The bseflow seprtion method ws pplied for 10 yers of stremflow dt from 80 guging sttions in the Moselle bsin. A comprison of both methods for 40 subbsins (Fig. 2) hs been crried out. The vlues for k rnge from 0.87 to 0.97 which leds to vlues of K from 0.01 to 0.40. The results for 11 subbsins were excluded from the nlysis becuse of extremely lrge discrepncies between nd within the two methods, indicting tht the dischrge dt re not homogeneous or tht the HGs were not correctly estimted. In Fig. 3 the sctterplot of the dily recession constnts estimted by hydrogrph nlysis nd by use of regression eqution (4) re shown. The coefficient of explined vrince by regression (R 2 ) between the recession constnts estimted by the two methods is 44% nd smller thn the vlue of 66% presented by Demuth (1993) for the upper Rhine nd upper Dnube region. A clcultion of the regression coefficients using the vilble dt in the Moselle bsin led to new coefficients nd ni? 2 of 52%, indicting sptil dependency of the vlues used in eqution (4). CONCLUSIONS The outlined study is first ttempt for regionliztion of the recession constnt using multivrite regression technique. In regrd of the subjectivity in estimting
164 Peter Krhe et l. the recession constnts by hydrogrph nlysis tking into ccount the smll scle of the input mps for outlining of hydrogeologicl units, the results cll for further reserch in this field. Improvements re expected () using geologicl nd hydrogeologicl mps with scle 1:500 000, (b) by the formultion of objective rules creting recession curves nd (c) by introducing t lest two prllel liner storges with different recession constnts to simulte bseflow. To tke into ccount the sptil dependency of the regression coefficients seprte dtsets of regression coefficients hve to be creted for different regions. Although ongoing studies re needed, the recession coefficients presently estimted by regression nlyses re in use for bseflow simultion in wter blnce nd precipittionrunoff models. The method described is lso helpful tool for testing the relibility of dily dischrge dt nd the results of hydrogrph nlyses in lrge river bsins. Acknowledgements The work presented in this pper ws supported in prt by Europen Environment Reserch Progrmme under Grnt no. EV5VCT930293 nd the Federl Ministry for the Environment. This reserch support is grtefully cknowledged. The uthors lso thnk Mr C. Frnzen for progrmming DEREC2 nd evluting the dily dischrge dt. REFERENCES CHR/KHR (1978) Ds Rheingebiet, Hydrologische Monogrphie/Le Bssin du Rhin, Monogrphie Hydrologique. Sttsuitgeverij Den Hg, ISBN 90127750. Commission of the Europen Communities (1985) Soil Mp of the Europen Communities 1:1000 000. Ctloguenumber CD4084A92END, Brussels Luxembourg. Demuth, S. (1993) Untersuchungen zum Niedrigwsser in WestEurop (Investigtion of low wter in Western Europe). Freiburger Schriften zur Hydrologie 1, Freiburg i. Br., Germny. Krhe, P. & Mendel, H. G. (1993) Verwendung von LndstTMDten zur meso und mkroskligen hydrologischen Modellierung (Use of LndstTM dt in meso nd mcro scle hydrologicl modelling). In: Tgungsbnd: 9. Nutzerseminr des Deutschen Fernerkundungsdtenzentrums der DLR. DLRMitt. 9308. Krhe, P., Dmen, K. & Pinnow, J. (1996) Continentl wter blnce computtion for Europe preliminry results. In: Second Interntionl Scientific Conference on the Globl Energy nd Wter Cycle. Preprint Volume. Wshington, DC. Thornthwite, C. W. & Mther, J. R. (1957) Instructions nd tbles for computing potentil évpotrnspirtion nd the wter blnce. Publictions in Climtology 10(3). Vorôsmrty, C. J. & Moore, III B. (1991) Modeling bsinscle hydrology in support of physicl climte nd globl biogeochemicl studies: An exmple using the Zmbezi River. Surveys in Geophysics 12, 271311, Kluwer Acdemic Publishers. Willmot, C. J., Rowe, C. M. & Mintz, Y. (1985) Climtology of the terrestril sesonl wter cycle. J. Climtology 5, 589606.