Figure
1 - Timeslice from a cross-spread land survey.
Target Finders have developed a specialised post-stack process to remove the acquisition footprint and condition 3D or 4D data for horizon time and amplitude picking, coherency and other attribute analyses. This will be illustrated using an example from an area suffering from high amplitude shot generated noise with a wide range of velocities. Despite an intensive prestack processing sequence, the stack volume contains a residual noise field with a spatial pattern relating to the acquisition geometry, as shown in Figure 1.
The noise field is organised according to the offset distribution in the CMP bins. Different offset distributions produce a different stack response and signal to noise ratio. The source lines run North-South on the plot and are spaced 600m apart, so the the offset distribution and the noise pattern cycle over 600m in the East-West direction. The source lines in this example are often deviated, which introduces some irregularities into the pattern. The receiver line component is not visible at this scale.
Figure
2 - TKxKy transform of the timeslice in Figure 1.
This TKK transform shows the periodic components in the timeslice. The 600m source line spacing produces the closely spaced spots at 1.67 cycles per 1000m in the Kx direction. The receiver line spacing of 200m produces the spots at 5 cycles per 1000m in the Ky direction. However, the noise pattern in the timeslices is not exactly periodic, due variations in the noise velocities and deviations in the source lines. There are subtle aperiodic components spread across the transform plane which cannot be attenuated by simple KK notch filtering.
Much of the signal component in Figure 2 is located close to KK=0, and overlaps the first noise sidelobe at Kx=+/- 1.67. However, other parts of the wavenumber plane contain signal which is invisible by eye. Simply notching out all the periodic noise spots in the KK domain will not only attenuate the signal around DC, but could attenuate signal with higher wavenumbers.
Figure 3 - Timeslice of starting noise estimate
The filtering technique is based on an image processing algorithm used to remove artifacts from optical scanner data. Although it originated as a 2D process, we have extended the algorithm to operate in three dimensions.
The first step is to obtain an estimate of the footprint. One way of doing this is to pass filter the periodic components in TKK space, and then transform back to TXY as shown in Figure 3. This is an enlargement of the footprint estimated from Figure 1, and shows the receiver line component running in the East-West direction as well as the source line component running North-South.
This noise field is used as the starting model for an iterative process which seeks to partition the input data into signal and noise, based on the statistics of the input and the current noise estimate. This process adapts the noise estimate to include random and partially organised noise which would not be rejected by KK notch filtering. The output from the process is simply the input minus the final noise estimate.
The benefits of this process over other spatial filtering techniques are as follows:
Since 1998 we have applied this process to thirteen 3D surveys, including what was in 1999 the largest continuously acquired land 3D in the World.
Horizon time and amplitude picking, attribute analyses and trace inversion all benefit from improved signal to noise ratio and a wavelet which more accurately reflects changes in acoustic impedance. Click here to see the results! The images may take a little while to download (138K).
Target Finders