International Conference on GPR in Archaeology (Nara, February 2001)

Processing and Visualisation of 4D-GPR Data

Immo Trinks

Derk Wachsmuth and Harald Stümpel

Time-lapse measurements of fluid movements in the subsurface become increasingly important not only in the exploration of hydrocarbon reservoirs but also in environmental geophysics. Rainwater oozing away in the subsurface does not penetrate as a uniform front but often follows preferential flow paths in form of so called "fingers". The GPR method offers advantages of comparatively fast detection of flow paths and fluid movements in soils with very high resolution (0.1-0.3 m). We demonstrate different processing and visualisation techniques for spatial and temporal GPR measurements.

During a project on preferential flow paths at the Institute of Geophysics and the Institute for Water Management and Landscape Ecology at the University of Kiel, Germany, a time-lapse infiltration experiment has been carried out in a full scale model. The aim of the presented experiment was to map water movements in the shallow subsurface in high spatial and temporal resolution under controlled circumstances.

The field laboratory consisted of a box (surface: 7 m x 5 m; base: 5 m x 3 m; depth: 2.0 m) filled homogeneously with sand. Tap water was pumped with an electrical pump from a water reservoir via a small plastic tube (Ø 1.1 cm) into the centre of the survey area. The tube was buried in the sand 5 cm below the surface. On an area 3 m x 1.2 m wide 30 GPR profiles were measured with 500 and 900 MHz antennas (GSSI). Both antennas were mounted on a motor driven sledge and pulled with constant speed. Positioning of the antennas was via punctured tape measure and a photo-electric barrier. The sampling grid was 1 cm inline, 4 cm crossline and 470 samples in time with a sampling rate of 0.119 ns.

At the onset of the experiment a complete data volume consisting of 30 parallel profiles was recorded over the dry sand volume. Afterwards 11 further data sets were measured sequentially without delay with a steadily increasing amount of infiltrated water. The registration of one data set required in average 23 minutes. Within 5 hours approximately 65 litres of water were infiltrated. We present the processing and analysis of the 500 MHz data set.

The first step of data analysis was a basic velocity analysis. Next, we investigated the reflectivity of the data sets by summing over the samples of the Hilbert transformed traces.

The damping effect of higher conductivity in the subsurface caused by the infiltration of water should affect the spectral content of the data. Therefore we analysed the integral of the quadratic spectral density. This procedure corresponds to a summation of the areas under the spectra of each GPR trace. An obvious coherence between fluid quantity and the quadratic spectral density is observed. The lateral extend of the region of higher soil moistness can be determined.

Time lapse measurements are undertaken to detect temporal changes in the subsurface. Assuming a fixed soil matrix such changes are based on the migration of fluids and gases. The main problem involved in investigating the same location at different times is the repeatability of the measurement. Differing measurement conditions can cause changes in the recorded signal which exceed the anomalies that we are looking for by several orders of magnitude. Unlike 4D seismic surveys that have time intervals between successive measurements in the order of month or years, our experiment had time intervals of hours and minutes (average 30 minutes). The accuracy in recovering measuring locations was very high (± 1cm). Vertical variations of the antenna positions were less than ± 1cm.

The changes in signal between two successive data sets can easily be computed through subtraction. These computations of temporal differences suppress features that don't change between successive measurements and therefore result in an improved signal-to-noise ratio.

2D migration was applied in order to improve the resolution of the sections. A 3D migration of the whole data volume is difficult to handle because of the lateral and vertical varying velocities within the region of the infiltrated fluid.

After subtracting from each GPR section the corresponding dry profile, GIF-images were created for each data set and merged into 3D cubes. The selection of different threshold values for the amplitudes allows insight into the data volume. By setting the low amplitudes of the dry sand transparent the flow path remains visible.

The creation of animations of the data volumes in form of slide shows enables the interpreter to scroll through the data volumes either in inline or crossline direction or in time. Furthermore, the animated infiltration process can be viewed in 2D as well as in 3D.

Animations of the presented data can be viewed under http://bullard.esc.cam.ac.uk/~trinks/radar.html

Return to conference page