Incorporating experimentally derived streamflow contributions into model parameterization to improve discharge prediction

Abstract

Abstract. Environmental tracers have been used to separate streamflow components for many years. They allow us to quantify the contribution of water originating from different sources, such as direct runoff from precipitation, subsurface storm flow, or groundwater to total streamflow at variable flow conditions. Although previous studies have explored the value of incorporating experimentally derived fractions of event and pre-event water into hydrological models, a thorough analysis of the value of incorporating hydrograph-separation-derived information on multiple streamflow components at varying flow conditions into model parameter estimation has not yet been performed. This study explores the value of such information to achieve more realistic simulations of catchment discharge. We use a modified version of the process-oriented HBV model that simulates catchment discharge through the interplay of hillslope, riparian-zone discharge, and groundwater discharge at a small forested catchment which is located in the mountainous north of South Korea, subject to a monsoon season between June and August. Applying a Monte-Carlo-based parameter estimation scheme and the Kling–Gupta efficiency (KGE) to compare discharge observations and simulations across two seasons (2013 and 2014), we show that the model is able to provide accurate simulations of catchment discharge (KGE ≥ 0.8) but fails to provide robust predictions and realistic estimates of the contribution of the different streamflow components. Using a simple framework that compares simulated and observed contributions of hillslope, riparian zone, and groundwater to total discharge during two sub-periods, we show that the precision of simulated streamflow components can be increased, while remaining with accurate discharge simulations. We further show that the additional information increases the identifiability of all model parameters and results in more robust predictions. Our study shows how tracer-derived information on streamflow contributions can be used to improve the simulation and predictions of streamflow at the catchment scale without adding additional complexity to the model. The complementary use of temporally resolved observations of streamflow components and modeling provides a promising direction to improve discharge prediction by representing model internal dynamics more realistically.