Theory

Figure 1 shows the typical flow for ray-based reflection tomography.

• First a set of reflectors are picked. Then residual moveout analysis is performed along the reflectors.
• This moveout is characterized by one or more parameters.
• The reflectors are also used as the basis for back projection points. The normal of the picked reflectors is used either for calculating the initial takeoff angle (if shooting up from the reflector) or for ensuring Snell's law is obeyed at the reflector (if shooting down to the reflector).
• Finally, time errors are estimated for each ray pair Stork (1992).

Two of the largest potential sources of error in this estimation scheme are an inaccurate dip estimate (causing information to back projected into the wrong portion of model space) and inaccurate description of moveout. The first type of error is the result of a poor description of the reflectors, which may be the result of overuse of an auto-picker or too little or too much smoothing of reflector positions. Poor moveout description is often the result of extending reflectors into areas with low signal-noise ratio where moveout analysis gives unreliable information. Both problems can be attenuating with significant human QCing, but will increase turnaround time.

 

old Figure 1 A typical ray-based reflection tomography loop.

Figure 2 shows an alternate ray-based tomography flow. Rather than using picked reflectors as the basis for back projection locations, points are selected according to reliability factors. First dip and coherency of the migrated image is calculated at each image location. For an initial dip and coherence estimate I take a window around each model location. I calculate the best single dip within the region, and the coherence of that dip, using the method described in Claerbout (1992). I then use this as an initial dip estimate for the non-linear, space varying dip estimation procedure described in Fomel (2000).

 

new Figure 2 Back-projection scheme used in this paper. Note the absence of reflector picking.

Likely back projection points are then automatically selected by finding model locations that meet some specified dip coherence, amplitude, and distance from other selected points. To get the `best' points in each region, these criteria are slowly relaxed (e.g. the first pass might look for points above the 90th percentile in amplitude and dip coherence, while the last pass might drop both these criteria to the 50th percentile.).

At each initially selected point semblance analysis is performed. Points that don't have good semblance (large semblance value and a definite maximum) are discarded. The remaining points are then used.