Discrete element modelling (DEM) is a numerical method for examining the dynamic behaviour of granular media. Soil is a nonuniform, discontinuity, elastic-plastic complicated medium. Although there are some advantages of using continuum numerical methods, the assumption of continuity is not always valid, as when there is a change in the soil structure and soil translocation. In order to overcome the shortcomings of the continuum numerical methods, discrete element modelling (DEM), which is a discontinuum numerical method for modelling the mechanical behaviour of granular materials, is a relevant approach for modelling soil as it can consider soil failure, deformation and translocation. In DEM simulation, soil is composed of individual particles which can be defined to be very small in size. And one of the key aspects for successful DEM simulation is to define and calibrate the model soil particles so that they reflect the behavior of real soil. This study presents a systematic method for calibrating a granular soil model, which is based on the surrogate model theory. This systematic method including four steps: (1) construct the soil contact model, the soil parameters were determined by two ways,measurement of three of the real-material properties through experiments and reference some soil parameters;(2) determination of the design variables defining the virtual soil, the design variables were selected by soil parameter sensitivity analysis;(3) construction of surrogate models for the virtual-material properties as a function of the design variables via simulated experiments;(4) optimization of the design-variable values to fit the virtual-soil properties to the real-soil values. The Edinburgh Elasto-Plastic Cohesion Model (ECM) was selected to fit the soil particle contact model, which was the elastic-plastic mechanics model. And the ECM contact parameters and it’s value were constant adhesive pull-off force f0：-0.001N， adhesive parameter stiffness kadh：45kN/m， tangential stiffness factor γt： 0.05，loading spring stiffness k1：100kN/m， unloading/reloading spring stiffness k2：613kN/m. The virtual soil model sensitivity analysis results showed that the particle radium, particle contact static friction and rolling friction should be calibrated and optimized. Finally, this paper constructed a soil model, the real soil sample experiment tested were taken from Zhuozhou city Hebei province, the soil texture is sandy loam soil. By this way, the optimized soil model parameters were particle radium 5.7mm，the static friction between two soil particles 0.45，the rolling friction between two soil particles 0.21, respectively. The soil penetration resistance tests simulation with optimized physical parameters were carried out. The real test about soil penetration resistance were carried out. Results showed that the difference values between virtual simulation and real test within 5.1%. And, it proved that the soil model after optimized could replace the real soil and take some simulation relate to soil-implement. This study provide the theory for soil model parameter calibration and optimization based on DEM method.