Towards reliable uncertainty quantification for hydrologic predictions, part II: Characterizing impacts of uncertain factors through an iterative factorial data assimilation framework

Abstract

In this study, an iterative factorial data assimilation (IFDA) framework is developed to holistically characterize the individual and interactive effects of various uncertain factors on hydrological predictions. The IFDA framework is flexible and is able to reveal the impacts from different numbers of uncertain factors. An iterative factorial analysis (IFA) approach is proposed in IFDA to diminish the biased variance estimation in traditional multilevel factorial designs and provide more reliable impact characterization for the considered factors. The proposed IFDA framework is applied to quantitatively reveal the individual and interactive effects of hydrological models, data assimilation (DA) methods, and uncertainties in inputs, streamflow observations and sample sizes on the deterministic and probabilistic predictions from data assimilation. The results indicate that the hydrological models, DA methods and their interactions would have the most dominant effects on hydrological predictions. This implies that different hydrological models or DA methods would produce significantly distinguishable results. When the hydrological model and DA method have been specified, uncertainties in streamflow observations would more likely have a visible effect on the accuracy of resulting predictions. Moreover, the inherent randomness, mainly caused by the Monte Carlo sampling procedures in data assimilation, would also have noticeable effects on the DA performances, especially when the hydrological model and DA method have been pre-identified. These results suggest that enhancement of hydrological models and data assimilation methods would be the most efficient pathway to generate reliable hydrological predictions.