GPR data time varying deconvolution by kurtosis maximization

Stochastic and deterministic deconvolution methods encounter difficulties in increasing the temporal resolution of GPR data. Statistical approaches, such as predictive or spiking deconvolution are not effective when the wavelet is not minimum phase, which is the case for GPR data. Wavelet deconvolution is not successful because the shape of the GPR wavelet changes with time. Here, prior to deconvolution, we apply a spectral balancing method in time–frequency (t–f) domain which efficiently produces GPR traces whose dominant frequency does not depend on time. We correct for phase residuals using the maximum kurtosis method. The methodology is demonstrated on synthetic and real GPR data.

► Spectral balancing in t–f domain reduces the non-stationarity of GPR traces.
► Spectral balancing is a precondition for wavelet deconvolution of GPR traces.
► Maximum kurtosis method obtains the phase residuals of the deconvolved GPR trace.