Prediction and subtraction of coherent noise using a data driven time shift: A case study using field 2D and 3D GPR data

• We remove diffracted air waves from 2D and 3D GPR data.
• The noise trajectory is picked and the data are flattened on that trajectory.
• The average wavelet in a moving window centered on the picked trajectory is noise.
• The noise is subtracted from the window's center trace and the data are unflattened.
• Amplitude and phase of the noise are implicitly included in data adaptive procedure.

We demonstrate the effectiveness of a methodology for reducing coherent noise in GPR data and show 2D and 3D GPR field data examples of subtracting direct and offline diffracted air waves. The procedure is guided by picking the noise time trajectory and shifting each trace to flatten on the picked trajectory. Subtraction of the coherent noise is implemented and optimized by data-adaptive wavelet estimation in a local time window that is centered on the flattened trajectory and moves across the traces. After the subtraction, the flattening is undone to shift each trace back to its original time. This procedure removes the coherent noise while retaining most of the primary reflected signals.