Feature-based Groundwater Hydrograph Clustering Using Unsupervised Self-Organizing Map-Ensembles

Abstract Hydrograph clustering helps to identify dynamic patterns within aquifers systems, an important foundation of characterizing groundwater systems and their influences, which is necessary effectively manage resources. We develope unsupervised modeling approach characterize cluster hydrographs on regional scale according dynamics. apply feature-based improve the exploitation heterogeneous datasets, explore usefulness existing features propose new specifically useful describe hydrographs. The itself based a powerful combination Self-Organizing Maps with modified DS2L-Algorithm, automatically derives number but also allows influence level detail clustering. further develop framework that combines these methods ensemble modeling, internal validation indices, resampling consensus voting finally obtain robust result remove arbitrariness from feature selection process. Further we measure sort clusters, for both interpretability visualization. test weekly data Upper Rhine Graben System, using more than 1800 period 30 years (1986-2016). results show our adaptively capable identifying homogeneous groups hydrograph resulting clusters spatially known unknown patterns, some correspond clearly external controlling factors, such as intensive management in northern part area. This easily transferable other regions and, by adapting describing features, time series-clustering applications.