In the context of railways, it is essential to have a reliable ground model to ensure their safety and smooth operation. To achieve this, geophysical monitoring, and geotechnical techniques, are often used. However, effectively combining geophysical and geotechnical data remains a challenge. In this study, a methodology based on data fusion of Ground Penetrating Radar (GPR) and Pandoscope data and based on belief functions is proposed in order to define the ground model of the railway substructure. The fusion of these data using Smets, Dempster-Shafer, and PCR6 normalization rules allows to improve the characterization of horizontal interfaces and to achieve an accurate assessment of the subsurface conditions on railway tracks.