Multiple-point geostatistical reconstruction of GPR reflection data

A common challenge in reflection GPR data processing and analysis is the reconstruction of missing traces. Gap filling, for example, may be needed to fill-in data where they could not be recorded in the field in order to produce a uniform trace spacing that is important for Fourier- or finite-difference-based migration methods. Similarly, field GPR data recorded in continuous mode with an uneven trace spacing are usually needed at a regular spacing for subsequent visualization and imaging. Finally, we may wish to increase the spatial resolution of a GPR dataset through “super-resolution”, whereby new traces are simulated between the existing ones in order to improve the interpretability of the data. A common challenge in these various applications is the need to interpolate a variable that has a complex, non-smooth behavior.

A number of interpolation methods have been proposed for filling in missing GPR traces over the past decades. The majority of these, however, tend to produce overly smooth and unrealistic results. Here, we present a data reconstruction strategy based on the QuickSampling (QS) multiple-point geostatistical method. With this approach, GPR traces are simulated via sequential conditional simulation based on patterns that are observed in nearby high-resolution data (training images). To evaluate the potential of our approach, we apply it to a variety of field 2D GPR datasets. Results indicate that the QS method provides an effective means of simulating missing GPR traces in a highly realistic manner.