From Streamflow Regime Identification to Regime Prediction in Ungauged Basins: Catchment Classification Across Türkiye

Hydrological systems are undergoing significant changes due to the combined effects of climate change, land surface alterations, and increasing human pressures. As a result of these influences, streamflow regimes show variations across temporal and spatial scales. Understanding the spatial and temporal variability of streamflow regimes and how they are affected by climate, anthropogenic factors and catchment characteristics are crucial for improving water resource management and for addressing challenges in ungauged basins.

This study presents a catchment classification framework based on the analysis of streamflow regime variability across Türkiye and applies a multi-method framework using observed discharge records obtained from 214 gauging stations. Streamflow regimes were identified using two different approaches and for this purpose the Agglomerative Hierarchical Clustering algorithm was utilized. The first approach is Functional Data Clustering with B-spline representations of monthly streamflow records. This approach enables the identification of streamflow regimes through hydrograph shape. The second approach utilizes Hydrologic Index-Based Clustering and uses a set of monthly flow indices. The resulting regime classifications obtained from both methodologies were compared to evaluate consistency, reliability, and hydrological interpretability of detected streamflow regimes across Türkiye.

To understand spatial and temporal variability in streamflow regimes, regime classification framework was first applied to the full observation period (1997–2015) to characterize long-term regime behavior and then subsequently to overlapping five-year sub-periods derived using a moving-window approach. This approach enabled the detection of potential streamflow regime shifts over time across Türkiye. The identified regime types were then correlated with an integrated dataset of climate indices, catchment attributes, land cover, soil type, and geology to investigate the controls on streamflow regime variability. Finally, a Random Forest Classification framework was used to assess the relative importance of multiple drivers and to enable the prediction of streamflow regimes in ungauged basins.

The results reveal that streamflow regimes have spatially distinct patterns and temporal variability across Türkiye. The findings also emphasize the critical role of elevation and precipitation seasonality in this variability. The consistency between the two classification approaches further supports the reliability of the identified regimes. Overall, the integrated framework combining streamflow regime classification, detection of regime shifts, and data-driven approach provides a basis for understanding streamflow variability and its dominant controls across hydrologically diverse and ungauged basins.