Möglichkeiten der Aggregierung heterogener Eingangsdaten für eine prozessorientierte hydrologische Simulation der Wasserflüsse am Beispiel des Untersuchungsgebietes der Oberen Leine

Abstract

Die für eine physikalisch basierte Simulation der Wasserflüsse benötigten Eingangsdaten und Modellparameter liegen naturgemäß in einer hohen räumlichen Heterogenität vor. Am Beispiel des mittelgroßen Einzugsgebiet der oberen Leine (ca. 1000 km²) werden verschiedene Methoden der Aggregierung verglichen und für eine Vorhersage der langfristigen Wasserbilanz erprobt. Besonderer Augenmerk lag dabei auf der Aggregierung der Bodeninformationen und der Landnutzungsparameter.<br /> In einer weiteren Untersuchung wurden verschiedene räumliche Diskretisierungseinheiten unterschiedlicher Größe für die Simulation der regionalen Wasserflüsse verwendet. Hierbei erzielen sowohl die Simulationen auf rasterförmigen Diskretisierungseinheiten als auch die auf polygonförmigen, an den Eingangsdaten orientierten Einheiten ähnlich gute Vorhersageergebnisse. Es zeigt sich dabei, dass die Güte und Auflösung der Niederschlagsdaten einen entscheidenen Einfluss auf die Verhersagegüte besitzt. Höhere Auflösungen können diese Güte nur noch geringfügig verbessern. Der Einfluss der Aggregierungsmethoden auf die Wasserbilanz erweist sich dabei als relativ gering.

The aim of this study is the development of methods for the aggregation of heterogeneous input data for process based hydrological models. The uncertainties of these methods with respect to different parameter sets are analysed. The most suitable methods were applied to a simulation of the water fluxes within the mesoscaled catchment of the upper Leine river (989 km²) with a physically based and distributed model. <br /> For comparison the catchment is unitised in grid cells as well as polygon shaped units (homogeneous areas of one sort of input data). In both cases a significant amount of subgrid variability in the input data is preserved. The remaining variability is then aggregated with the developed methods. <br /> All the input data are available in different spatial resolutions which results in several variabilities. Due to the fact that the data of the given meteorological input are rather homogeneous compared to the other input data, normally no aggregation is neccassary for these data. Instead a disaggregation is carried out with respect to the relief information. In contrast, land use data, relief data and soil data are more variable in space, so these data have to be aggregated. In the present study an aggregation of the evapotranspiration fluxes is performed to aggregate the land use data. The complete heterogeneous structure can be taken into account within the simulations. This means that each model unit is subdivided into subareas of homogeneous land use. On each subarea the relief data are averaged in a preprocessing and then evapotranspiration is calculated. After the calculation of each subarea within the model unit the aggregation of the fluxes (i.e. an area weighted arithmetical average) is done in each time step. The soil water fluxes are calculated once per simulation unit using the aggregated evapotranspiration. For the model unit which has got the highest heterogeneity in land use data, this flux aggregation predicts the averaged results of a fully distributed simulation of the model unit. The results are better than the results by using the majority land use within a simulation. Nevertheless there are some uncertainties related to the flux aggregation because of nonlinearities in the reduction of the potential evapotranspiration. The simulated runoff components are influenced by these uncertainties more than other water fluxes. <br /> The aggregation of the heterogeneous soil data consists of two different problems. On one hand the vertical structure of the soil profiles has to be considered, and on the other the soil parameters have to be aggregated. The total soil volume within a simulation unit is shown to have a significant influence on the model results. It therefore should be preserved within an aggregation method. The pronounced aggregation onto the effective thickness of the soil profiles preserves the total soil volume and results in slight underestimations of the average of the nonaggregated soil profiles. Two methods of aggregating the soil profiles are compared. Both of them provide acceptable results. For the aggregation of the soil parameters five methods are investigated. The results of all of them are quite good, but according to all circumstances to be studied one method can be trusted slightly more than the others. A combination of geometrical and arithmetical mean with respect to the distribution function of the parameter provides the best results. In most instances an aggregation of already pedotransferred soil parameters is to be preferred to a primarily aggregation of soil properties followed by a pedotranfer function. <br /> All the developed aggregation methods are used for the calculation of the water fluxes within the catchment. A total number of 20 various scenarios was simulated. The scenarios differ either in the model units or in the aggregated parameters used in the simulation. All of them are quite similar for the temporal prediction of the water fluxes but vary much for the spatial prediction. The temporal predictions are as well in a good correspondence with the measurement for the runoff at different gauges all over the catchment. Beneath a certain spatial scale (i.e. the scale of the homogeneous precipitation areas of about 75,6 km² as the mean) of the model units, the correspondence does not arise in a significant further amount. The grid sizes choosen in this study (500m, 1000m and 2000m) are all smaler than this scale. The results according to the simulation of the different grid sizes are quite similar. <br /> Nevertheless the uncertainty increases with the grid size. With respect to the applied simulation model it can be concluted that the uncertainties are acceptable when calculating the water fluxes on greater grid sizes. This is in fact helpful for the coupling of hydrological and meteorological simulation models. <br /> As a result the applied aggregation of the soil parameters brings about more uncertainties than the applied aggregation of the land use data. On the basis of the given runoff measurements none of the investigated scenarios could be selected for the best calculation of the water fluxes for sure. The lack of precise spatial data for the validation of the model results makes a final assessment impossible. To surmount this lack will be an important task for future investigations. Data for several properties within a high spatial resolution are most desirable.