Spectral bandwidth enhancement of GPR profiling data using multiple-frequency compositing

Abstract The amplitude spectrum of ground penetrating radar (GPR) reflection data acquired with a particular antenna set is normally concentrated over a spectral bandwidth of a single octave, limiting the resolving power of the GPR wavelet. Where variously-sized GPR targets are located at numerous depths in the ground, it is often necessary to acquire several profiles of GPR data using antennas of different nominal frequencies. The most complete understanding of the subsurface is obtained when those frequency-limited radargrams are jointly interpreted, since each frequency yields a particular response to subsurface reflectivity. The application of deconvolution to GPR data could improve image quality, but is often hindered by limited spectral bandwidth. We present multiple-frequency compositing as a means of combining data from several frequency-limited datasets and improving the spectral bandwidth of the GPR profile. A multiple-frequency composite is built by summing together a number of spatially-coincident radargrams, each acquired with antennae of different centre frequency. The goal of the compositing process is therefore to produce a composite radargram with balanced contributions from frequency-limited radargrams and obtain a composite wavelet that has properties approximating a delta function (i.e. short in duration and having a broad, uniform spectral bandwidth). A synthetic investigation of the compositing process was performed using Berlage wavelets as proxies for GPR source pulses. This investigation suggests that a balanced, broad bandwidth, effective source pulse is obtained by a compositing process that equalises the spectral maxima of frequency-limited wavelets prior to summation into the composite. The compositing of real GPR data was examined using a set of 225, 450 and 900 MHz GPR common offset profiles acquired at a site on the Waterloo Moraine in Ontario, Canada. The most successful compositing strategy involved derivation of scaling factors from a time-variant least squares analysis of the amplitude spectra of each frequency-limited dataset. Contributions to the composite from each nominal acquisition frequency are clear, and the trace averaged amplitude spectrum of the corresponding composite is broadened uniformly over a bandwidth approaching two-octaves. Improvements to wavelet resolution are clear when a composite radargram is treated with a spiking deconvolution algorithm. Such improvement suggests that multiple-frequency compositing is a useful imaging tool, and a promising foundation for improving deconvolution of GPR data.