<p>Reliable deformation analysis of rockfill dams is essential for ensuring deformation safety throughout the entire lifecycle. Model parameter updating is a critical method for improving the accuracy of deformation analysis for rockfill dams. Although geotechnical parameters usually exhibit certain correlations, previous studies often assume that individual parameters follow a normal distribution or neglect prior knowledge. This can lead to the low fidelity of the updated parameters and poor accuracy in deformation analysis, further limiting its application. Thus, this study introduces a knowledge-guided approach for parameter updating of deformation analysis models for rockfill dams by incorporating parameter dependencies as prior knowledge. This scheme leverages copula-based estimation of distribution algorithms (EDAs) along with ensemble learning surrogate models to determine the optimal parameters. Specifically, a copula model was developed to capture the multivariate joint distribution of the model parameters based on the 511 sets of parameters collected from 48 rockfill dams. By sampling the copula model, the initial population is created and iteratively refined by incorporating high-fidelity individuals. Two sets of experiments demonstrated that the parameters updated with the proposed method achieved high fidelity and accuracy. The application results in a 219-m-high concrete-faced rockfill dam showed that the average relative error in deformation analysis based on updated parameters was 6.8%. The knowledge-guided method is demonstrated to reliably and stably support parameter updating, accurately reflecting the nonlinear evolution and spatial distribution patterns of dam deformation.</p>