Parameter sensitivity of automated baseflow separation for snowmelt-dominated watersheds and new filtering procedure for determining end of snowmelt period

Automation in baseflow separation procedures allowed fast and convenient baseflow and baseflow index (BF and BFI) estimation for studies including multiple watersheds and covering large spatio-temporal scales. While most of the existing algorithms are developed and tested extensively for rainfall- and baseflow-dominated systems, little attention is paid on their suitability for snowmelt-dominated systems. Current publishing practice in regional-scale studies is to omit BF and BFI uncertainty evaluation or sensitivity analysis. Instead, “standard” and “previously recommended” parameterizations are transferred from rainfall/BF to snowmelt-dominated systems. We believe that this practice should be abandoned. First, we demonstrate explicitly that the three most popular heuristic automated BF separation methods—Lyne–Hollick and Eckhardt recursive digital filters, and the U.K. Institute of Hydrology smoothed minima method—produce a wide range of annual BF and BFI estimates due to parameter sensitivity during the annual snowmelt period. Then, we propose a solution for cases when BF and BFI calibration is not possible, namely excluding the snowmelt-dominated period from the analysis. We developed an automated filtering procedure, which divides the hydrograph into pre-snowbelt, post-snowmelt, and snowmelt periods. The filter was tested successfully on 218 continuous water years of daily streamflow data for four snowmelt-dominated headwater watersheds located in Wyoming (60–837 km ² ). The post-snowmelt BF and BFI metric can be used for characterizing summer low-flows for snowmelt-dominated systems. Our results show that post-snowmelt BF and BFI sensitivity to filter parameterization is reduced compared with the sensitivity of annual BF and BFI and is similar to the sensitivity levels for rainfall/baseflow systems.