When should stream water be sampled to be most informative for event-based, multi-criteria model calibration?

Abstract

Abstract Isotope data from streamflow samples taken during rainfall or snowmelt events can be useful for model calibration, particularly to improve model consistency and to reduce parameter uncertainty. To reduce the costs associated with stream water sampling, it is important to choose sampling times with a high information content. We used the Birkenes model and synthetic rainfall, streamflow and isotope data to explore how many samples are needed to obtain a certain model fit and which sampling times are most informative for model calibration. Our results for nine model parameterizations and three events, representing different streamflow behaviours (e.g., fast and slow response, with and without overflow), show that the simulation performance of models calibrated with isotope data from two selected samples was comparable to simulations based on isotope data for all 100 time steps. Generally, samples taken on the falling limb were most informative for model calibration, although the exact timing of the most informative samples was dependent on the runoff response. Samples taken on the rising limb and at peakflow were less informative than expected. These model results highlight the value of a limited number of stream water samples and provide guidance for cost-effective event-based sampling strategies for model calibration.