An important prerequisite for the generation of realistic material behavior with the Discrete Element Method (DEM) is the correct determination of the material-specific simulation parameters. Usually, this is done in a process called calibration. One main disadvantage of classical calibration is the fact that it is a non-learning approach. This means the knowledge about the functional relationship between parameters and simulation responses does not evolve over time, and the number of necessary simulations per calibration sequence respectively per investigated material stays the same. To overcome these shortcomings, a new method called Metamodel-based Global Calibration (MBGC) is introduced. Instead of performing expensive simulation runs taking several minutes to hours of time, MBGC uses a metamodel which can be computed in fractions of a second to search for an optimal parameter set. The metamodel was trained with data from several hundred simulation runs and is able to predict simulation responses in dependence of a given parameter set with very high accuracy. To ensure usability for the calibration of a wide variety of bulk materials, the variance of particle size distributions (PSD) is included in the metamodel via parametric PSD-functions, whose parameters serve as additional input values for the metamodel.