Importance of the informative content in the study area when regionalising rainfall-runoff model parameters: the role of nested catchments and gauging station density

Abstract

The setup of a rainfall-runoff model in a river section where no streamflow measurements are available for its calibration is one of the key research activities for the Prediction in Ungauged Basins (PUB): in order to do so it is possible to estimate the model parameters based on the hydrometric information available in the region. The informative content of the dataset (i.e. which and how many gauged river stations are available) plays an essential role in the assessment of the best regionalisation method. This study analyses how the performances of regionalisation approaches are influenced by the “information richness” of the available regional dataset, i.e. the availability of potential donors, and in particular by the gauging density and by the presence of nested donor catchments, which are expected to be hydrologically very similar to the target section. The research is carried out over a densely gauged dataset covering the Austrian country, applying two rainfall-runoff models and different regionalisation approaches. The regionalisation techniques are first implemented using all the gauged basins in the dataset as potential donors and then re-applied, decreasing the informative content of the dataset. The effect of excluding nested basins and the status of “nestedness” is identified based on the position of the closing section along the river or the percentage of shared drainage area. Moreover, the impact of reducing station density on regionalisation performance is analysed. The results show that the predictive accuracy of parameter regionalisation techniques strongly depends on the informative content of the dataset of available donor catchments. The “output-averaging” approaches, which exploit the information of more than one donor basin and preserve the correlation structure of the parameter, seem to be preferable for regionalisation purposes in both data-poor and data-rich regions. Moreover, with the use of an optimised set of catchment descriptors as a similarity measure, rather than the simple geographical distance, results are more robust to the deterioration of the informative content of the set of donors.