What is the best time to take stream isotope samples for event-based model calibration?

Abstract

Environmental tracer data, such as the stable water isotopic composition of streamwater and precipitation, are valuable for understanding runoff generation processes and calibrating hydrological models. Despite recent technical advancements, the collection and analysis of streamwater and precipitation samples still involve significant costs and efforts. Consequently, it is useful to study how many samples need to be collected as a basis for model calibration and what the most informative sampling times are. In previous studies, we used the Birkenes hydrological model and synthetic data to explore the value of a few stream isotope samples for event-based model calibration. Our results showed that the information from two or three selected isotope samples, and particularly a sample taken during the falling limb of the hydrograph, improved model performance significantly compared to calibration against streamflow data only. In this follow-up study, we used a unique Swiss dataset with high-frequency isotope measurements of precipitation and streamwater during six rainfall-runoff events to determine which stream isotope samples were most informative for model calibration. Several benchmarks were used to judge model validation performance. Our results showed that for 83% of the 42 possible combinations of calibration and validation events two strategically selected streamwater samples improved model performance compared to the lower benchmark when only stream stage data were used for calibration. The most informative samples also improved the calibration of a mixing-related parameter. When the model was calibrated and validated for the same event, the most informative samples for model calibration were those from the early part of the falling limb, which confirmed the results of the previous studies with synthetic data. Pre-event isotope samples were more informative when the model was calibrated for one event and then validated against other events or all events combined. Our results suggest that relatively inexpensive event-based data for model calibration can be obtained for so-far ungauged catchments by placing a water level logger in the stream and collecting precipitation isotope data and a few streamflow samples. These samples should preferably include a pre-event sample and a sample from the early falling limb. However, similar analyses for a wider range of sites and events, in combination with a selection of models with different structures and assumptions, are needed to confirm these results and to optimise sampling strategies further.