DREAM(LoAX): Simultaneous Calibration and Diagnosis for Tracer-Aided Ecohydrological Models Under the Equifinality Thesis

Abstract

The Limits of Acceptability approach has been demonstrated to be an effective conditioning tool due to its capacity to consider epistemic uncertainty. However, its application faces two challenges—the low efficiency when random sampling is used and the difficulty in setting limits prior to calibration. Here an algorithm DREAM_(LoAX) was developed and added to GLUE framework. As an extension of DREAM_(LoA) of Vrugt and Beven (2018), https://doi.org/10.1016/j.jhydrol.2018.02.026, it evaluates model performance based on limit boundaries, thus inherits the merits of the GLUE framework (explicit consideration of epistemic errors). Moreover, the importance of initial choice of limits is strongly reduced by allowing iterative evolution of limits based on historical model performance. By testing a series of examples (including a high-dimensional numeric example, a single-objective hydrological example, and a multi-objective hydrological example) with or without error-free assumption using synthetic or real observations, the search capacity of DREAM_(LoAX) to locate acceptable models is demonstrated. The algorithm also shows comparable efficiency to DREAM and DREAM_(LoA). More importantly, it provides real-time diagnostic information regarding when (at which timestep), where (for which objective), and how (to which direction and to which extent) the model fails when uncertainty is pronounced, allowing potential uncertainty sources in the data or flaws in the model structure to be identified. In this context, DREAM_(LoAX) is not only a useful conditioning tool, but also a diagnostic and learning tool for development of improved modeling.