Stimulated by the inquiry of using high frequency electromagnetic ground waves to communicate information among unattended ground sensors, numerical simulations of GPR performance in different near-surface geological settings were conducted. Two sets of 400 MHz GPR field data, one from Fort Richardson, Alaska, and the other one from Hanover, New Hampshire, were used to be the 'ground truth' to compare with numerical simulations. The numerical simulation algorithm we used adapts the finite difference time domain method, with a perfectly matched layer as the absorption boundary condition to truncate outbound waves. The signal impulse has a central frequency of 400 MHz, and the time step is 0.067 ns. We have simulated four cases: a combination of two radiation polarizations (TM and TE), and two geological settings, i.e., a sandy/gravelly half-space overlain by a silty/clayey layer (the case of Fort Richardson, AK), and a silty/clayey half-space overlain by sandy/gravelly layer (the case of Hanover, NH). The results depict the following implication. (1) More EM energy is radiated into the air as an air wave for the TM mode, and more EM energy will be sent into the ground when the TE mode is used, regardless of the geological setting. (2) Where a gravelly sandy half-space overlain by a silty/clayey layer, more EM energy will be trapped in the silty/clayey layer as a ground wave guide in the TE mode with almost no air radiation, when compared with the same radiation mode in the case of silty/clayey half-space overlain by a layer of gravelly sandy. (3) For the geological setting of a sandy/gravelly half-space overlain by a silty/clayey layer, the TE mode only contains ground wave and the TM mode only contains air wave energy. This implies that for this case a far more complete separation of the air wave and the ground wave can be reached. These simulation results imply that transmission mode should consider the on-site geological setting when attempt to use the ground wave as a communication carrier.