Abstract: The paper presents a novel approach to enhancing the accuracy of turbulence modeling in nuclear reactor simulations, specifically addressing secondary flow in the fuel assembly of pressurized water reactors. The accurate prediction of secondary flow in the rod bundle channels of pressurized water reactors is crucial for the design and performance evaluation of nuclear fuel assemblies. Traditional numerical simulation methods have difficulty in striking a balance between computational cost and prediction accuracy. This paper addressed this challenge by calibrating the constants of the nonlinear eddy viscosity model (NLEVM) based on the high-fidelity flow field measurement data from a 5×5 rod bundle with a split mixing vaned spacer grid using a novel data assimilation strategy that incorporated the ensemble Kalman filter (EnKF) algorithm. This study enhanced the overall data assimilation strategy and the EnKF algorithm by introducing sensitivity-based deterministic sampling and correlation-length-based local adjustment, respectively. These modifications aimed to accelerate convergence and reduce residuals in the model constant calibration process. Besides, this study adopted the correlation coefficient between the predicted and measured flow fields as a criterion for judging whether the calibrated model constants improve the prediction accuracy of the secondary flow in the rod bundle channel, which is more in line with the actual phenomenon than the traditional relative error criterion. The calibrated NLEVM significantly improves the prediction accuracy of secondary flow in the rod bundle channels compared to the standard NLEVM. The similarity between the predicted flow field of the calibrated model and the experimentally observed flow field is improved, with the correlation coefficients of the full cross-section flow field improving to a greater extent the further away from the localized grids. The secondary flow structure of the subchannels predicted by the calibrated model agrees well with the experimental observations and can successfully predict the cross-sectional vortex structure that is not accurately predicted by the original model. These results demonstrate effectiveness in refining turbulence model predictions through the novel data assimilation strategy and the modified EnKF algorithm. This research represents a substantial contribution to computational fluid dynamics, particularly in the context of nuclear reactor applications. The innovative approach to calibrating turbulence models using data assimilation strategies paves the way for more accurate and reliable predictions of turbulent flows in complex geometries, with broad-reaching implications for various scientific and engineering disciplines. The study’s findings could be instrumental in enhancing the safety and efficiency of nuclear reactor operations and potentially applicable in other fields requiring precise turbulence modeling, such as aerospace engineering, climate modeling, and industrial process optimization. The integration of data assimilation strategies with traditional turbulence models opens new avenues for improving the fidelity of simulations in complex flow scenarios.