Abstract: Current dynamic analyses of earth-rock dams are hindered by the lack of reliable criteria for determining the stochastic nature of dynamic parameters, resulting in limited research on the influence of spatial variability of rockfill materials on dam dynamic response. Based on field measurement data from an engineering project, this study extracted statistical characteristics (mean values and coefficients of variation) of physical parameters of rockfill materials. By utilizing the Geo-Gravel database for coarse-grained soil dynamic properties, combined with random field theory and the Seft-Net deep learning model, a nonlinear mapping relationship between the random field of physical parameters and the random field of dynamic parameters for rockfill materials was established. By comparing the results given by finite element analysis of deterministic parameters, the influence of spatial variability of rockfill materials on the dam′s dynamic response was investigated. The results indicated that when considering the stochastic nature of field-measured physical parameters, the maximum horizontal displacement and the peak acceleration at the dam crest increased by 14.8% and 9.2%, respectively, compared to the deterministic calculation values. Furthermore, the degree of underestimation of extreme dynamic response values of the dam increases with higher coefficients of variation.