Yes ! It is possible to process GPR data wit SU to use all the power of seismic processing tools to GPR data.

The basic trick is to fool SU by using micro second as the time unit and centimeter instead of meter. This way :

• frequencies are in MHz
• velocities range from 3330 cm/micro sec (water) to 30000 cm/microsec (air)
• offset keyword value is not badly truncated

Note : this is sufficient for low frequency GPR used for geological applications (25-1000 MHz) but may be adapted for higher frequency GPR.

There is no unified GPR data format and each manufacturer produces its own files. GPR data need to be imported from the manufacturer's format to SU format using the right tool.

• For Sensor&Software devices you can use the dt1tosu tool included in SU. (test and sample data needed)
• For Malå devices you can download a shell script here.It is included in SU44R15 and higher.
• For other manufacturers, ask for the detailed format information and share the tool !

Once the data are converted to SU format, you will probably need to adjust the geometry using tools like sushw and/or suchw

Compared to seismic data (apart from the recording or sampling time and size of the survey) GPR data exhibit a few differences.

• Absolute propagation time is generally unknown. Actually, GPR propagation is so fast that the receiver can't start diitizin when receiving a trigger signal from the transmitter, like in seimics. The source is fired at regular intervals and the digitizer is shifted by one sample at each pulse. The absolute time is not known but can be deduced when one knows the distance between antennas. Consequently you have to apply a static correction to GPR signal using sustatic to exhibit real travel time.
• GPR signal is generally strongly polarized and affected by some very low frequencies caused by electronic noise. This is easily corrected by removing the mean of the trace (sugain mbal=1) and applying a high pass filter consistent with the antennas band-pass
• The direct wave in the air (even with shielded antennas) and in the ground is very strong. It often saturates the analog to digital converter. They must be muted prior applying bandpass filters. This can be done using sumute mode=1 or the dedicated sugprfb tool.
• The intrinsic attenuation of GPR signal is much faster (with respect to the wavelength) that for seismic signal and require the early application of a strong gain (with sugain or other techniques).

Like in seismics, there is no such thing as standard processing. Processing aims at mitigating noises that limit the usability of the data and at enhancing primary reflected signal. This is always a tradeoff between enhancements and negative side effects which heavily depends on the data.

The dominant frequencies of the antennas are well known in the air, but for geological applications, the influence of the ground can dramatically change the useful frequencies of the signal. The best way is to locate the low and high frequencies where the recorded signal contains no or little geologic information by low or high pass filtering the data.

Sometimes, GPR signal is affected by ringing. For single trace processing, you can try predictive deconvolution (supef) or if the ringing is laterally consistent a median filter with a null moveout (sumedian median=1 xshift=0 tshift=0).

Once the signal is clean, if the goal of the survey is not to evidence diffraction hyperbolas, it is often fine to migrate the data. GPR velocities tend to vary less that seismic velocities et a simple Stolt migration (sustolt) with a constant velocity is often enough. In normal GPR configuration the normal move out is quickly null. In order to use zero offset migration, it may be necessary to perform a normal moveout correction (sunmo) to correct the time of the first nanoseconds.

A time to depth correction (suttoz) gives a touch of finish to your data. Of course if you are using a constant velocity, it can be replaced by the right d1 value in the trace headers. Since d1 (and d2) trace header words are floats, you can use them to label you data in meters instead of centimeters.

— Dominique ROUSSET 2019/07/25 12:43 dominique