Contribution of model parameter uncertainty to future hydrological projections

Abstract. Hydrologic models have been applied to predict land surface water and energy budgets in mountainous watersheds that are characterized by complex geological features and climatic variability. A common practice is to calibrate the models and achieve the best performing parameter set according to historical observations, and then the calibrated model was used to do future projections. One drawback is that the influence of parameter uncertainty on model projections is not well discussed. In this study, we applied multiple objective functions to choose a group of best performing parameter sets to the Boulder Creek Watershed, USA to investigate how parameter uncertainties can propagate to future projections. We used 16 parameter sets that have similar performance in simulating streamflow amount and regime historically, and applied the same parameter sets to predict hydrologic variables including streamflow, evapotranspiration, and soil moisture in two future phases (Phase 1 is 2040-2069 and Phase 2 is 2070-2099). The results show that variability due to parameter uncertainty was up to 10 % annually and 26 % monthly under future climate change scenarios, and the uncertainties are especially prominent during May to September. The different parameter sets can result to annual streamflow changes in opposite directions. The results indicate that a single parameter set may yield biased representation of hydrologic variability. It is necessary to consider multiple optimal parameter sets in applying hydrologic models for hydrological projections and water resources decision making.