A newly proposed data assimilation framework to enhance predictions for reflood tests

[Display omitted] •The higher acceptance rates and faster computational performance.•A new accuracy evaluation and Monte Carlo sampling approach are proposed.•The better minimum and more stable convergence of the system state were also observed compared with PAPIRUS. Data assimilation is the process that can enhance the predictions by adjusting the input parameters inside their uncertainties range. In this study, the predicted results of the reflood tests, which were simulated by the SPACE code, were examined using the data assimilation technique. Crucially, a newly proposed data assimilation framework, STARU (Sampling meThod for highly non-lineAR system Uncertainty analysis), was developed. To propose the new data assimilation framework, we combined a new accuracy evaluation method with the Monte Carlo sampling algorithm to enhance the predictions for such a complex system as reflooding phenomena (many input parameters and non-linear responses). Subsequently, the outcomes of STARU were compared with the performances of data assimilation in PAPIRUS, a toolkit with many functions and modules for uncertainty analysis, which was developed by Heo and Kim (2015). Conversely, STARU only focused on solving problems with many parameters and highly non-linear systems in which implementation of the Monte Carlo sampling method is required because of the complexity of the systems. The main objectives of developing STARU are to reduce the computation time and increase the acceptance rate of the data assimilation process for complex systems. Consequently, we found that STARU effectively enhanced the data assimilation results for the reflood tests with higher acceptance rates and faster convergence. Furthermore, better improvements and more stable convergence of the system states were also observed, which effectively facilitated the search process for the most sensitive physical models in the predictions. In the future, STARU may be suitable for implementation in the other complex problems to calibrate the models, examine the uncertainty propagation, and study the sensitivity analysis.