Fig. 1
Valentina Krysanova, Dirk-Ingmar Müller-Wohlfeil, Alfred
Becker
Water quality modelling for mesoscale watersheds represents a field where a compromise solution between very complex models with many parameters and inadequate simplified models should be found. The paper includes a description of a new watershed model SWIM which integrates hydrology and water quality. It is based on two previously developed models - SWAT and MATSALU - and is intended for application in mesoscale watersheds with an area of up to 20,000 km2. SWIM includes some modules common to both models, trying to combine their advantages and to avoid overparametrization. A SWAT/GRASS interface is one of the advantages of SWAT, and it is adapted for the use in SWIM. The method of spatial disaggregation is more advanced in MATSALU, and thus it is implemented for SWIM. The model runs under the UNIX environment. Simultaneously to the development of the model, it is being tested in subbasins of the Elbe drainage basin.
Watersheds are important as integrators of many forces, including climate. Their natural boundaries and hierarchical structure represent an appropriate structure for environmental impact analysis and modelling. At present, knowledge about the dynamics of dominant processes in large watersheds is still rather limited due to the extremely complicated character of these processes and the interrelation of many factors of different nature (Beven, 1995). Reliable assessment of nonpoint source (NPS) pollution is one of the problems involved, especially for larger watersheds.
Watershed simulation models represent physical and biochemical processes in a dynamic way. Conceptually, such models describe mathematically water fluxes and associated pollutant fluxes from the land surface and soil profile. Source areas can be categorized in accordance with a distinct land use/land cover and soil type. Dissolved and solid-phase concentrations of chemical compounds can be obtained from lumped modelling of biogeochemical cycling at a source area. These concentrations vary with land cover, soil type, management practices and season of year. Transport (or retention) factors reflect a complex chain of physical and biochemical processes which can affect nutrient movement from a subbasin to the river outlet and must be taken into account.
The complexity of the specific watershed simulation model depends on the temporal and spatial resolution, and on the extent to which important biochemical processes are considered in the model. While there is a certain progress in water runoff simulation for larger basins, erosion and pollutant transport from larger watersheds represents a field where a compromise solution between very complex models with too many parameters and inadequate simplified models should be found.
Previous efforts in watershed modelling were concentrated mainly on developing either continuous-time spatially lumped models or single event spatially distributed models. Most of the models tended to focus on the patch scale or small homogeneous watersheds, where data availability is certainly better (models CREAMS (Knisel, 1980), EPIC (Williams et al., 1984), OPUS (Smith, 1992)). Recent development of deterministic models provides also spatially distributed tools, among them: AGNPS (Young et al., 1989), ANSWERS (Beasley et al., 1980), SWRRB (Arnold et al., 1990), MATSALU (Krysanova et al., 1989a & b), SWAT (Arnold et al., 1993). In a sense, all the models made use of previous approaches. In particular, CREAMS and its components were used as a basis for many further tools. For example, SWRRB is a distributed version of CREAMS, which can be applied to a basin with a maximum of 10 subbasins, and SWAT is an extended and improved version of SWRRB, running simultaneously in several hundred subbasins. However, the general tendency is that the data requirements increase exponentially with the increase in watershed size.
The availability of GIS tools and more powerful computing facilities made it possible to overcome many difficulties and limitations and to develop distributed continuous time models, based on available regional information. While the application of AGNPS and ANSWERS is limited to watersheds of about 200 km2, SWRRB was developed with limited distributed parameter capability to be used in agricultural basins as large as 600-800 km2, MATSALU was applied in a 3,500 km2 rural basin, and SWAT is intended to be applied in watersheds up to 25,000 km2. The SWAT represents a component of the HUMUS project, where it is applied for 350 6-digit hydrologic unit areas in the 18 major river basins in the U.S. (Srinivasan et al., 1993b) All these models are to a certain extent integrated with GIS tools.
In this paper a new watershed model SWIM, which integrates hydrology and water quality, is presented. This model is based on SWAT and MATSALU, and is adapted for the use in European conditions. Simultaneously with development, the model is applied for subbasins of the Elbe drainage basin.
The German part of the Elbe river drainage basin (about 96,000 km2) has been chosen for our regional study due to several reasons. Firstly, it is one of Europe`s largest river basins, situated in Central Europe (80% of the area belong to the former GDR), and sharing similarity in climate and data availability with other European rivers. Secondly, the basin with predominately sandy soils is exposed to comparatively low amount of precipitation, and is characterized by high water demand (both climatic and anthropogenic), all these factors predetermine its high hydrological vulnerability. Thirdly, the Elbe is one of the most heavily contaminated water courses in Europe, due to ineffective sewage water treatment and lack of nonpoint source pollution control (agricultural areas cover about 56% of the total area). The new model is intended to be applicable to other river basins in Europe. To assure this, the input data and parameters are preferably taken from the public domain sources and treated in as universal manner as possible.
The new model SWIM (Soil and Water Integrated Model) is a simulation continuous-time spatially distributed watershed model, based on two previously developed models:
SWAT - (Arnold, Allen, Bernhardt, Srinivasan, Muttiah, Walker, Dyke, 1993, USDA & Texas A&M University), and
MATSALU (Krysanova, Meiner, Roosaare, Vasilyev, 1989, Estonian Ac. Sci.).
SWAT is a continuous-time distributed simulation watershed model. It was developed to predict the effects of alternative management decisions on water, sediment, and chemical yields with reasonable accuracy for ungauged rural basins. The model was developed by modifying the SWRRB (Arnold et al, 1990) and ROTO (Arnold, 1990) models for application to large, complex rural basins. Major changes include: (a) expanding the model to allow simultaneous computations on several hundred subbasins, (b) adding components to simulate groundwater flow, routing transmission losses, and sediment and chemical movement through ponds, reservoirs, and streams. The model operates on a daily time step. A discretization scheme allows subdivision into cells and/or subbasins. SWAT is integrated with the GIS GRASS and a relational data base to extract necessary input parameters.
MATSALU is a system of four simulation models for a mesoscale agricultural watershed and the ecosystem of a sea bay. It was developed for and applied in the Matsalu Bay watershed in Estonia in order to evaluate different management scenarios for eutrophication control of the Bay. Spatial disaggregation is based on the overlay of three maps (elementary subwatersheds with an average area of 10 km2, land use, and soil) to obtain so-called Elementary Areas of Pollution (EAP). The model includes four coupled submodels. At first, water balance is calculated for each EAP with a daily time step to define soil moisture and runoff components (modification of the SCS CN method). After that nutrient balance in soil and nutrient losses from the EAP are estimated. Then water and nutrients are routed in stream flow, and, finally, nutrient cycling in the sea bay ecosystem is calculated. While originally the model was not integrated with a GIS, recently some efforts were made in order to provide integration of chemical submodel with ARC/INFO (Meiner, 1995). Nevertheless, the model is not sufficiently transferable to be used directly in other watersheds.
Why was it necessary to develop a new version of a watershed model? The reason is that no one of the existing models can be used directly for any European watershed larger than 1,000 km2. While AGNPS has a limited scale of application, MATSALU is not sufficiently transferable, and SWAT is currently adapted only for US watersheds (using the specific data sets, particularly soil and weather data bases) and is tested with the monthly time step as a long-term predictor. On the other hand, practically all the mentioned models are in development, none of them is perfect.
For example, it is clear that the current scheme of spatial disaggregation in SWAT (a basin is subdivided into subbasins, and then the dominant soil and land use are used to characterize this subbasin as the lumped parameters) could be and should be improved. So, as an initial step, the spatial disaggregation scheme from MATSALU was introduced into SWAT. The next step was to adjust the model for the use in European conditions, where data availability is different. Currently the model SWIM includes some common modules of the both predecessors: the SCS CN method for surface runoff, the MUSLE approach for erosion, a simplified EPIC approach for crop growth, trying to combine their advantages: the three-level spatial disaggregation scheme and nutrient modules from MATSALU; the GRASS interface, hydrological modules and routing procedure from SWAT, and to avoid over-parametrization.
SWIM integrates weather, hydrology, erosion, crop growth, and nutrients (nitrogen and phosphorus) at the watershed scale. The following hydrological components are included:
| component | method |
|---|---|
| precipitation | real input data or weather generator |
| snow melt | f(water content of snow, daily max temp.) |
| evapotranspiration | Pristley-Taylor method for PET, Richie's model for AET |
| surface runoff | modified SCS CN method |
| subsurface runoff | storage routing technique for soil |
| percolation to ground water | storage routing technique for soil |
| ground water flow | f(ground water recharge) |
The following chemical components are included:
| component | method |
|---|---|
| fertilisation | input data (agriculture statistics) |
| mineralization | f(org. matter, temp., soil water) |
| plant consumption | f(crop growth, crop type) |
| denitrification | f(nitrate-N, soil water) |
| leaching into ground water | f(g-w recharge, soluble nutrient content) |
| sorption/desorption for P | f(soil type, P content) |
| transport with surface flow | f(soluble nutrient content, surface runoff) |
| transport with subsurface flow | f(soluble nutrient content, subsurface runoff) |
Transport (or retention) factors are taken into account through averaging (weighted average) of water and nutrient fluxes for heterogeneous subbasins and routing of water, sediments and nutrients in the river flow through transmission losses. The model is intended to be applied in mesoscale watersheds with an area up to 20,000 km2.
If we want to have an efficient model of water quality for the regional scale, an appropriate balance in the model parametrization has to be found. On the one hand, the model has to be detailed enough to account for a large diversity of processes involved. On the other hand, if the model is fully deterministic, its use for large watersheds is limited. Current model development tries to overcome these difficulties. The neural network to extract channel geometry in SWAT, and wide use of weather generators for water quality modelling are only two examples. We suggest implementing one else stochastic procedure in regional-scale models, namely the stochastic allocation of crops for agriculture areas.
Really, it is extremely difficult to obtain current data on crop distribution in some tens to hundreds of subbasins, at least for areas with wide varietys of crops and crop rotations. In our opinion, some stochastic crop generators should be used for that in parallel with existing weather generators, especially for larger basins. The idea is that a limited number of crop rotation schemes can be defined for a certain region, based on expert knowledge. After that the crop generator is applied to distribute the crops in a watershed in accordance with land use, soil, and "crop probability" in the region. This procedure is intended to be included in SWIM.
Several water quality models, including AGNPS, ANSWERS, SWRRB, and SWAT, have been integrated with a GIS (most often ARC/INFO or GRASS) to facilitate the use of spatially distributed data. One major difference in extracting inputs for distributed watershed models is the method of disaggregation to drive the input parameters and submodels. Models like AGNPS or ANSWERS divide the study area into square grids, extract inputs for each grid cell, and apply the model to every cell.
In models like SWRRB or SWAT, a basin is divided into subbasins based on elevation only or on elevation and homogeneity of soil and crop. After that, tools are provided to aggregate inputs at both subbasin level (to extract either the mode or weighted average characteristics for soil and crop in a subbasin) and basin level (to evaluate water and chemical routing and transmission losses) for the model. A default procedure in SWAT is used to aggregate the soil series categories and land use for each subbasin using the mode (dominant) aggregation method.
In the MATSALU model the whole basin (about 3,500 km2) was divided into subbasins of major tributaries, which, in turn, were subdivided into so-called Elementary Watersheds (EW) with an average area of 10 km2. A map of elementary watersheds existed prior to the study. After that, an overlay of land use and soil maps onto the EW map created the Elementary Areas of Pollution - homogeneous plots of land with uniform land cover and soil type in the EW. Such a three-level disaggregation has proven to be quite satisfactory.
According to our experience, spatial disaggregation into elementary units based on natural features (topography, land use, soil types, ground water table) is preferable in comparison with disaggregation based on square grids, as this essentially reduces the number of elementary units and computing time while preserving the accuracy of calculations.
A three-level disaggregation is implemented in the SWIM model for meso-scale basins. It is more complicated than that in SWAT, but it seems to be more reliable. The idea is that a mesoscale basin with an area of 50 to 20,000 km2 is first subdivided into subbasins of reasonable average area (see explanation below). It can be easily done using the r.watershed program of GRASS (or any other GIS with similar capabilities), which is applied to a DEM (Digital Elevation Model) of the whole area. Then hydrotops or elementary units are delineated within every subbasin, based on land use and soil types. Normally, a hydrotop is a set of disconnected units in the subbasin. The three-level disaggregation implies 1) basin, 2) subbasins, and 3) hydrotops.
As an example, Fig. 1 demonstrates the scheme of spatial disaggregation for the Elbe drainage basin. We do not intend to apply the model for such a large basin as a whole. That is why, the whole watershed (Fig. 1, e) should be first subdivided into subwatersheds of a reasonable average area. In the case of Elbe the whole drainage basin (96,000 km2) was subdivided into 57 subwatersheds with an average area of 1,700 km2 (Fig.1, d) using r.watershed function in GRASS. Then the three-level spatial disaggregation scheme is applied to these subwatersheds (which are later called "basins") (Fig.1, c) - into subbasins (Fig.1, b) and hydrotops inside the subbasins (Fig.1, b and a). After this disaggregation of the basin, an aggregation procedure is used for modelling, starting from the hydrotop level up to the basin level, as indicated by arrows at the Fig. 1. After the verification of results for the Elbe subwatersheds (Fig.1, d), the routing scheme can be applied to combine the outputs for the whole Elbe watershed.
Fig. 1
The SWAT/GRASS interface includes some very useful operations (for example, to extract the routing structure and the weather stations), and it is partially used for SWIM. At the same time, some additional operations were needed to implement a new spatial disaggregation scheme. For example, the SWIM/GRASS interface creates a "structure file" to drive the model. Every line in the file describes the characteristics of one hydrotop - its land use, soil, and subbasin number.
A very important question is how to choose reasonable spatial resolution? This problem is of fundamental significance for hydrological and hydrochemical process modelling. According to Kuchment (1922), the level of detail adopted in representing the spatial variability of vertical moisture fluxes (which is assumed to be about 100m) depends on the ranges of horizontal mixing in the surface turbulent boundary layer (which should be of the order of 10 times more than this depth). The conclusion follows that it is sufficient to take into account only heterogeneities on scales of 1 km or larger. This means that grid cell size (to be combined into hydrotops) should be not larger than 1 km.
According to Beven and Kirkby (1979), the effect of the channel network probably becomes important for basins larger than about 10 km2, where the time constant of the network (i.e. travel time through it) becomes as long as for the infiltration phase. And it is also known that 10 km (or 100 km2) appears to be the minimum scale, beyond which inhomogeneities in land surface properties can trigger specific meso-scale atmospheric circulation systems, which have a definite impact on land surface - atmosphere interactions (Kuchment, 1992). So, an average subwatershed area, where the effect of the river network can be neglected, should be not more than 10 - 100 km2.
The restriction on average subwatershed area and time step influence the computing time, and, taken together with data availability, define the upper limit of watershed area for the model application. Such reasonable spatial disaggregation should allow the applicability of the model to be extended to larger basins.
The model operates on a daily time step. The SWAT/GRASS interface (Srinivasan, Arnold, 1993, Srinivasan et al., 1993a) is adopted for SWIM to extract spatially distributed parameters of elevation, land use, soil types, and groundwater table. It is modified where necessary as described below. The interface creates a number of input files for the basin and subbasins, including the hydrotop structure file and the subbasin routing structure file. Steps 1, 2, 5, 6 are used unchanged, steps 3, 4 are new, some other steps from SWAT/GRASS (such as irrigation and nutrient attributes) are excluded.
1. Subbasin attributes. This is the first step to be fulfilled. Using a given subbasin map, the program calculates area, resolution, and coordinate boundaries for the basin and each subbasin. The fraction of each subbasin area to the basin area is calculated.
2. Topographic attributes. The program estimates the stream length, stream slope and geometrical dimensions, accumulation area, and aspect. The weighted average method is used to estimate the overland slope and slope length. Finally, the channel factors K and C of the Universal Soil Loss Equation (USLE) are estimated using a standard table.
3. Hydrotop structure. The program defines the basin structure by overlaying the subbasin map with land use and soil layers. The structure file is created to run the model. Every line in the file describes the characteristics of one hydrotop - its land use, soil, and subbasin number.
4. Weather attributes. The program selects the closest weather station to the subbasin. Then either actual weather information, or weather generator (in the case the long-term monthly statistics are available for precipitation and temperature for the station) can be used.
5. Ground water attributes. The ground water parameters are estimated for each subbasin using the alpha layer, which defines the time lag needed to the groundwater flow as it leaves the shallow aquifer to return to the stream (Arnold et al., 1993).
6. Routing structure. This very important step in the SWAT/GRASS interface creates the routing structure for subbasins, based on the elevation map. Also, it defines the channel width and depth using a neural network that is embedded in the interface, based on the drainage area and average elevation of a subbasin.
After that the relational meteorological data, crop data base, soil data base, data on point sources of pollution and river routing parameters are read from the files. Additionally, river discharge and concentrations of nutrients are needed for model verification. Currently the model is tested using actual weather data.
The digital soil map of Germany (Bodenübersichtskarte der Bundesrepublik Deutschland 1 : 1 000 000) provides information for 72 soil types. Each soil type is characterized through a "leading profile". For each horizon of every soil profile, 8 attributes are specified: depth, texture class, clay content, humus content, carbon content, nitrogen content, field capacity, available field capacity. Based on the map legends, a special soil data base was created, which is used in SWIM to extract or estimate necessary soil parameters. For example, saturated conductivity is estimated based on the method described in OPUS (Smith, 1992), and soil erodibility factor is estimated from soil texture classes.
After the input parameters are read from files, the three-step modelling procedure is applied:
1. water and nutrient balance are calculated for every hydrotop (Fig. 1 a),
2. the outputs from hydrotops are averaged (weighted average) to estimate the subbasin output, not accounting for lag time in the case of surface runoff, and taking the average for subbasin lag time for subsurface flow (Fig. 1 b),
3. routing procedure is applied to the subbasin outputs, taking into account transmission losses (Fig. 1 c).
In parallel to the model development, a first application is performed for the Buckener Au catchment (64 km2 - a subbasin of the river Stör and the Dahme basin (up to Markisch Buchholz, about 535 km2) - subbasin of the river Spree (both Stör and Spree are tributaries of the Elbe). These rivers are not regulated and it enables us to test hydrological components with daily time resolution.
Here the results for hydrological cycle modelling are demonstrated for the Buckener Au watershed, which was subdivided into 9 subbasins and 72 hydrotops. Simulation was performed for 4 subsequent years 1989 - 1992.
Time series of different water flow components: surface runof (qd), subsurface flow (ssf) and percolation (perc) in 1990 are shown in Fig. 2 for three hydrotops - cropland on soil 25, forest on soil 25, and pasture on soil 6. The soil 6 is Niedermoorboden in German classification, which corresponds to Eutric Histosols in the FAO classification, the soil 25 is Podsol-Parabraunerde / Podsol-Fahlerde in German classification, which corresponds to Spodic Luvisols/Spodic Podzoluvisols in the FAO classification. The hydrotop "pasture on soil 6" can be considered as a wetland, which is dominated by organic soils. Here, percolation is most intensive (Fig. 2), whereas the surface runoff is highest in the cropland and lowest in the forest.
Fig. 2
The dynamics of the soil water index (soil water content / field capacity) for the same three hydrotops is shown in Fig. 3. In the hydrotop "pasture" soil is almost saturated even in summer time, while in the cropland and forest hydrotops the soil moisture is decreasing down to the wilting point in August.
Fig. 3
Fig. 4 demonstrates the averaged water flows for subbasin 2, where the abovementioned hydrotops dominate and occupy 69% of the whole area: cropland on soil 25 covers 25%, forest on soil 25 covers 28%, pasture on soil 6 covers 16% of the subbasin area.
Fig. 4
Fig. 5 shows the time series of evapotranspiration and soil moisture index for the subbasin 2.
Fig. 5
The precipitation and observed versus simulated runoff in the river outlet are demonstrated in Fig. 6 for two years - 1990 and 1992.
Fig. 6
The results obtained for hydrological components in Buckener Au and Dahme watersheds are satisfactory. The dynamics of water flows in hydrotops and subbasins is quite reasonable, and the Nash-Sutcliff (1979) efficiency is about 0.68 - 0.72 for these four years of simulation with daily time step. The next steps are to test the erosion, crop, and nutrient components.
After verification of results for some representative subwatersheds (Fig. 1 d), the model can be applied to other subwatersheds, and then a number of methods can be used to obtain results for the whole Elbe, like
- integration of modelling results for subwatersheds by means of a river routing model,
- nested watershed approach (upscaling based on smaller scale applications),
- application of a simplified model based on Unit Area Load estimates (which can be obtained from the modelling results).
A new version of watershed model integrating hydrology and water quality is presented in the paper. The model is based on existing tools and tries to combine their advantages and avoid overparametrization. A three-level scheme of spatial disaggregation implemented in SWIM for mesoscale basins seems to be quite reliable. Currently the model is tested in the Elbe subbasins. The intention is to extend the applicability of the model to other river basins in Europe.
We are very grateful R.Srinivasan (Blackland Research Center, Temple, TX) and J. Arnold (USDA ARS, Temple, TX) who kindly provided the SWAT codes and were always open for fruitful discussion.
The authors would like to thank Prof. Ripl (Free University of Berlin) and the Projektzentrum Ökosystemforschung, the University of the Armed Forces, Munich, and the German Weather Service (DWD) for providing data.
We are grateful to the Deutsche Forschungsgemeinschaft (DFG) and the Federal Ministry for Education, Science, Research and Technology (BMBF), which made the present work possible by providing financial support.
Arnold J.G. 1990. ROTO - a continuous water and sediment routing model. ASCE Proc. of the Watershed Management Symposium. Durango, Co, p. 480-488.
Arnold J.G., Allen P.M., Bernhardt G., 1993. A comprehensive surface-groundwater flow model. Journal of Hydrology, 142, 47-69
Beasley D.B., L.F.Huggins, E.J.Monke, 1980. ANSWERS: A model for watershed planning. Transactions of the ASAE. 23(4): 938-944
Beven K.J., M.J. Kirkby, 1979. A physically based, variable contributing area model of basin hydrology. Hydrological Sciences Bulletin, 24, 1, 3, 43-69.
Beven K. 1995. Linking parameters across scales: subgrid parametrizations and scale dependent hydrological models. Hydrol. Process., 6, 279-298.
Knisel, W.G. (ed.), 1980. CREAMS: A field scale model for chemicals, runoff, and erosion from agricultural management systems. USDA, Conservation Research Report NO. 26, 643p.
Krysanova, V., Meiner, A., Roosaare, J., Vasilyev, A., 1989a. Simulation modelling of the coastal waters pollution from agricultural watershed. Ecological Modelling, 49, pp. 7-29.
Krysanova, V. and Luik, H., (eds.) 1989b. Simulation modelling of a system watershed - river - sea bay. Tallinn, Valgus, 428 (in Russian).
Kuchment L.S., 1992. The construction of continental scale models of the terrestrial hydrlogical cycle: ana analysis of the state-of-the-art and future prospects. J-P.O'Kane (ed.) Advances in Theoretical Hydrology, European Geophys. Soc. Series on Hydrol. Sciences, 1, ELSEVIER.
Meiner A., 1995. Integration of GIS and a dynamic spatially distributed model for non-point pollution management. Proc. of the Second Int. IAWQ Spec. Conf. and Symposia on Diffuse Pollution, Brno & Prague, Czech Republic, August 13-18, 1995.
Smith R.E..1992 Opus, An Integrated Simulation Model for Transport of Nonpoint-Source Pollutants at the Field Scale: Volume 1 & 2.U.S. Department of Agriculture, Agricultural Research Service, ARS-98.
Srinivasan, R., and Arnold, J.G., 1993. Basin scale water quality modelling using GIS. Proceedings, Applications of Advanced Inform. Technologies for Manag. of Nat. Res. June 17-19, Spokane, WA, USA.
Srinivasan, R., Arnold, J.G., Muttiah, R.S., Walker C., Dyke P.T., 1993. Hydrologic Unit Model for the United States (HUMUS). In: Sam S.Y.Wang (ed.) Advances in Hydro-Science and -Engineering, Vol. I.
USDA, 1992. STATSGO - State soils geographic data base. Soil Conservation Service, Publ. Number 1492, Washington D.C.
Williams J.R., K.G. Renard, P.T. Dyke, 1984. EPIC - a new model for assessing erosion's effect on soil productivity. Journal of Soil and Water Conservation 38(5): 381-383
Young R.A., C.A.Onstad, D.D.Bosch, W.P.Anderson, 1989. AGNPS: a nonpoint source pollution model for evaluating agricultural watersheds. J.Soil and Water Cons. 44(2): 168-173
Address for correspondence:
Valentina Krysanova, Dirk-Ingmar Müller-Wohlfeil, Alfred Becker
Potsdam Institute for Climate Impact Research
P.O.Box 601203, Telegrafenberg
14412 Potsdam, Germany
Telephone: +49-(0)331-288-2515
Fax +49-(0)331-288-2600
E-mail: valen@pik-potsdam.de