Transition of cluster membership of hydrological states in Austria

Europe and the alpine region are specifically affected by climate change, which does not only increase air temperature, but changes the seasonality of hydrological regimes. Common approaches to identify homogeneous hydrological regions may be misleading, as regions can change over time. Therefore this thesis aims to investigate the potential of transitions in cluster solutions over time. Daily discharge data from 310 Austrian gauging stations for the summer 2003 drought (1 June to 31 August) was used. A functional clustering was chosen to reduce dimensionality and retain the overall shape of the hydrograph. The coefficients of a B-spline basis were clustered by K-means, with the number of clusters determined by the Elbow method, Gap statistic and silhouette width for different distance metrics. Cluster transitions between two periods of 46 days, and between spring, summer and fall 2003, were analyzed with the MONIC framework, which distinguishes external transitions (survival, emergence, merging, splitting and disappearance) from internal shifts in the location or density of the clusters. The four identified clusters show an altitudinal gradient from high-alpine nival to lowland pluvial regimes. The stability analysis yields a high Adjusted Rand Index (0.90 to 0.94 across basis sizes, bootstrap median 0.88). Within the summer, three of the four clusters survived, while the mid-elevation cluster dissolved at every threshold tested. Between seasons, one of the four clusters survived from spring to summer (two merged, one dissolved) and two from summer to fall (two dissolved), while the 15 high-alpine stations stayed together throughout. A simulation study showed that the location statistic detects a known shift in 98% of 100 replicates at a false-positive rate of 3%. Within the drought the broad regime structure largely held while its surviving clusters moved, whereas between seasons the classification itself was reorganized. Overall, functional representations combined with cluster-transition analysis can identify both changes in regime membership and within-cluster changes that static classifications cannot capture, providing a sensitivity-based framework for monitoring hydrological regimes during extreme events.