Estimation of the reference hydrological conditions in Slovenia with application of clustering analysis

The reference hydrological conditions describe the natural discharge as it would be without the exploitation of the water resources. The measured values of the discharge are obtained in the scope of national hydrological monitoring and are in most cases reduced for the amount of abstracted water. Therefore, the values measured in this way provide information on the amount of residual water in the river and not the total amount of water that would be available without abstraction. However, knowing the natural hydrological state is important when calculating the ecological flow and planning the future water use. For this purpose, our goal was to established the methodology for estimating the reference discharge on any water body in Slovenia with catchment size larger than 10 km². The development of the methodology was based on detailed simulations of reference hydrological conditions for 56 selected cases. As those simulations require extensive data preparation and are very time consuming, we intended to generalize the results obtained for selected cases to the whole country using clustering analysis. The hierarchical clustering and K-means approach were applied taking into account different model arguments (e.g. number of clusters, distance metrics, number of iterations). First we have grouped the simulation points to check which of the attribute data influence the classification the most. Than clustering was repeated on the data set representing points distributed over the whole country as well as simulation points. However, the further analysis of the clustering results and application of other methods for generalization showed, that clustering analysis is in this case suitable for analysis of patterns in data and identification of influential variables, while generalization turned out to be better performed applying multiple regression analysis.