Catchment clustering in Taiwan through comparative assessment of baseflow separation methods

60 catchments, Taiwan This study presents catchment clustering based on reliable baseflow estimation, using long-term streamflow data (1949–2022) from widespread gauging stations across Taiwan. The reliable baseflow estimation is based on integrating the selection of seven separation methods (Lyne-Hollick, Chapman-Maxwell, Eckhardt, UKIH, Fixed, Sliding, and Local interval) and parameters. A sensitivity analysis was conducted on the filtering parameters, and an objective benchmark based on recession flow for baseflow estimation was employed to select optimal method–parameter combinations. In addition, assessing the influence of reservoirs on BFI ensures the representativeness of the selected catchments. Finally, catchments were classified using k-means clustering based on the seasonal BFI. The LH method (α=0.925) and the ECK method (α=0.925, BFI max =0.8) as optimal method–parameter combinations, achieving an NSE of 0.97 and 0.99, respectively. Most rivers in Taiwan are dominated by subsurface contributions, with about 60% of streamflow coming from subsurface sources especially in mountainous regions. However, hydrogeological maps indicate that these regions are generally unfavorable for substantial groundwater contributions. Moreover, BFI was found to be largely unaffected by hydraulic infrastructures. The 60 catchments were classified into 5 distinct groups, each characterized by unique BFI magnitudes and seasonal dynamics. These findings provide valuable insights for selecting separation method parameters, understanding the composition of streamflow and the spatial variability of catchment behaviors of each group in Taiwan. • Catchments were classified into five groups based on the seasonal BFI. • Lyne and Hollick (α = 0.925) and Eckhardt (α = 0.925 and BFI max = 0.8) are the optimal configurations for estimating baseflow. • Rivers in Taiwan are typically subsurface water-dependent (∼60% of total streamflow).