© 13th International Conference on Applications of Statistics and Probability in Civil Engineering, ICASP 2019. All rights reserved.A physical domain-based formulation of parametric reduced-order models based on dominant normal modes and interface reduction is presented in the context of uncertainty propagation analysis. Normal and interface modes are approximated in terms of a set of support points in the uncertain parameter space. Based on these approximate modes, reduced-order matrices can be updated efficiently during the simulation process associated with the uncertainty analysis. The proposed approach is applied to a complex nonlinear finite element model under a synthetic ground motion generated by a realistic source-based model for subduction mega-thrust earthquakes.