We are interested in Ground Penetrating Radar (GPR) as a geophysical tool useful for determining the depth of the ground water table (GWT) and for monitoring shallow water infiltration in sandy soils. At hydrostatic equilibrium, the water content distribution in a homogeneous unsaturated soil down to the water saturated zone depends on the soil water retention function. A classical way to fit retention curve data is to use the van Genuchten continuous model. Using Finite Difference Time Domain simulations, we study the sensitivity of the GPR signal reflected by a van Genuchten type transition to the hydraulic parameters. We show a power type relationship between the reflected signal amplitude and the slope of the soil retention curve. Furthermore, for simulating GPR reflection data acquired above a transition from unsaturated to saturated soil, geophysicists often approximate the soil water retention curve by a piecewise linear model. We test the validity of such an approximation depending on the frequency of the radar signal and the abruptness of the retention curve. We illustrate our results with high resolution GPR data (1600 MHz) acquired above a fluctuating water table in a sand column at the laboratory scale.