Characterizing Impact Factors on the Performance of Data Assimilation for Hydroclimatic Predictions through Multilevel Factorial Analysis

Abstract

In this study, the multi-level factorial analysis approach is employed to characterize the major impact factors on the performances of different data assimilation schemes. Four data assimilation methods, including EnKF and PF methods, and two integrated data assimilation methods are adopted for real-time hydrological prediction through a conceptual rainfall-runoff model in a catchment of Jing River. Different uncertainty scenarios for model inputs and outputs, as well as streamflow observations are tested through the multilevel factorial analysis to track the dominant impacts factors on the performances of data assimilation approaches. The multi-level factorial results suggest that, for different data assimilation schemes, the impacts from stochastic perturbations in model inputs, outputs and streamflow observations are different and some of them may be statistically insignificant. But the impact for one factor is generally dependent upon the others and scenarios with extreme stochastic perturbations (low or high) may more likely result in a good performance for all data assimilation schemes.

Keywords: data assimilation, ensemble kalman filte, particle filter, multi-level factorial analysis, uncertainty