Scattered ground roll is a type of noise observed in land seismic data that can be particularly difficult to suppress. Typically, this type of noise cannot be removed using conventional velocity-based filters. In this paper, we discuss a model-driven form of seismic interferometry that allows suppression of scattered ground-roll noise in land seismic data. The conventional cross-correlate and stack interferometry approach results in scattered noise estimates between two receiver locations (i.e. as if one of the receivers had been replaced by a source). For noise suppression, this requires that each source we wish to attenuate the noise from is co-located with a receiver. The model-driven form differs, as the use of a simple model in place of one of the inputs for interferometry allows the scattered noise estimate to be made between a source and a receiver. This allows the method to be more flexible, as co-location of sources and receivers is not required, and the method can be applied to data sets with a variety of different acquisition geometries. A simple plane-wave model is used, allowing the method to remain relatively data driven, with weighting factors for the plane waves determined using a least-squares solution. Using a number of both synthetic and real two-dimensional (2D) and three-dimensional (3D) land seismic data sets, we show that this model-driven approach provides effective results, allowing suppression of scattered ground-roll noise without having an adverse effect on the underlying signal.