Ground Penetrating Radar (GPR) has been widely used to detect soil structure, soil moisture content, and soil texture. In this article, we establish models of 3D stochastic media in GPR based on soil fractal characteristics and perform GPR forward modeling. We use a 3D Fractal Brownian Motion (FBM) spectral density function to simulate stochastic media and discuss the effects of Hurst exponent and scale coefficients on the modeling results. The Hurst exponent is an important indicator in FBM, which reflects the smoothness of the medium as well as the degree of disturbance. Hurst's exponent is also used to describe the self-similarity of the soil, i.e. the degree of fractal. Soil Water Content (SWC) is a key factor affecting the heterogeneity of soil media, and the distribution of soil dielectric properties also depends on SWC. Therefore, we use SWC data to obtain the Hurst exponent through Rescaled range (R/S) analysis. Finally, a 3D stochastic medium model is established based on actual SWC data, and GPR forward simulation is performed on the established model by the Finite-Difference Time-Domain (FDTD) method. The simulation results show that the proposed stochastic medium modeling based on soil fractal characteristics can provide an effective modeling tool for GPR soil detection.