In this paper, the time-frequency separation algorithm (TFS) proposed by Belouchrani and Amin is applied to ground penetrating radar (GPR) data to reduce ground clutter, that hides reflected waves from a near-surface planar interface. We formulated the problem with several assumptions so that narrow band signals, whose center frequency and baseband signal depend on propagation paths, are received at the receiver, when a wideband signal is radiated from a transmitter. These phenomena can be clearly seen in time-frequency distribution (TFD) of the received signal. In this paper, we adopted the TFS utilizing the TFD signature as a blind separation technique to separate the ground clutter from the target signals. We show numerical and experimental results in order to verify the validity of the problem formulation and the TFS. We carried out GPR measurements to measure permafrost in Yakutsk, Russia. We found the difference in TFD signatures between the ground clutter and the target signal in the experimental data. We could detect the upper boundary of the permafrost with the TFS in spite of the unstable ground clutter.