Modeling Result - 15Hz Zero-Phase Ricker Wavelet

Velocity model used in this case is the same as one used in 25Hz zero-phase ricker wavelet case. Only difference is the input source has principal frequency 15Hz now. Figure 1 is the waveform of actual wavelet and Figure 2 is the frequency spectrum.

Figure 1. 15Hz zero-phase ricker wavelet

Figure 2. 15Hz zero-phase ricker wavelet frequency spectrum

This time, 700 time steps are computed instead of 600 time steps in 25Hz case since corresponding seismic volume has time range 2900-3600ms. The seismogram volume has dimension now 207X303X700. This volume will be desampled in time first and direct wave will be removed from recorded seismogram. Then volume interpolation process will reinterpolate seismogram from 20x20 ft space to 41x65 ft space.

Computed Seismograms

Seismograms for 1988 and 1994 both have dimensions of 207x330x700.

Preprocess Modeled Seismic Amplitude

Since source function used to compute synthesized seismogram is not calibrated with seismic amplitude, we have to calibrate amplitude level of modeled seismogram and real seismic, this is done by using 4D module amplitude matcher which does global amplitude scaling and bandpassing. The ideal case will be extracting seismic wavelet from the seismic itself and used that to be source in forward modeling.

Figure 3 is frequency spectra and amplitude histogram of the modeled seismogram and real seismic volume of South timberwolf 1988 survey. The green color is modeled seismogram and red is real seismic.

Figure 3. Frequency spectra and amplitude histograms after global match for 1988 model

from the frequency spectra, we can see now frequency spetrum of seismogram and that of real seismic is very close now. Due to slower take-off of source used in the modeling, therefore we move up the modeled seismogram 10 samples (40ms) up in order to match real seismic and modeled seismogram.

Figure 4 is frequency spectra and amplitude histogram of the modeled seismogram and real seismic volume of South timberwolf 1994 survey. The green color is modeled seismogram and red is real seismic.

Figure 4. Frequency spectra and amplitude histograms after global match for 1994 model

from the frequency spectra, we can see now frequency spetrum of seismogram and that of real seismic is not that close like 1988 modeling result. Also we moved seismogram 10 samples (40ms) up in order to match real seismic and modeled seismogram.

Modeling and Real Seismic datasets for Comparison

1985

File norm_seismogram_41x65ft.fld is the globally normalized seismogram for 1988 model, bandpass is 0-35Hz and 40ms shifted up. seismogram_20x20ft.fld is original seismogram recorded by Dr. Teng's FEM program, but the time sampling interval is now 4ms instead of original 1ms. seismogram_41x65ft.fld is spatially interpolated of dataset seismogram_20x20ft.fld in order to compare with seismic grid 41x65ft. The 4D module volume interpolater is used to do interpolation.

1994

File norm_seismogram_41x65ft.fld is globally normalized of file seismogram_41x65ft.fld which is interpolated file seismogram_20x20ft.fld. Modeled seismogram is moved up 10 samples to account slower take-off of source function used in modeling.

Seismic data is only bandpassed by 0-35Hz bandwidth, input seismic data is data used in 4D impedance inversion .

Direct Comparison of modeled and real seismic

Since real seismic data is time migrated and modeled seismogram is plane-wave simulation which is approximation of zero-offset post-stack seismic data. We first compare them against each other to see if any similarity exists between them. Here we show a set of seismic section extracted from modeled seismogram and real seismic.

Figure 5. Line 1780 modeled and real seismic

Figure 6. Line 1825 modeled and real seismic

Figure 7. Line 1850 modeled and real seismic

Figure 8. Cross Line 3866 modeled and real seismic

Figure 9. Cross Line 3940 modeled and real seismic

Figure 10. Cross Line 3940 modeled and real seismic

Figure 11. Time slice at 1960ms modeled and real seismic

Figure 12. Time slice at 3000ms modeled and real seismic

Figure 13. Time slice at 3280ms modeled and real seismic

Figure 14. Time slice at 3280ms modeled and real seismic

In the overlayed wiggle plots, BLACK wiggle is modeled seismogram and RED is real seismic. We can see that modeled seismic matchs real seismic at the shallow very well (wiggle plot of time slice 2960 and 3000) but not that good at the deep part(time slice 3280 and 3400). Line and cross line section show similarity at the shallow and difference at the depp part.

From wiggle plot of inlines and cross lines, we can see freqquency of real seismic now is close to modeled seismic.

Boundary reflection suppression

From figure 5 to figure 10, we see boundary reflection contaminates modeled seismogram, major velocity layer in the model is almost dipping west, so boundary is dipping east, this east-dipping events can be suppressed by dipping F-K filtering. 4D module dip filter is used for this purpose. Figure 15 and 16 are parameters used in dip-filtering, the modeled seismogram is first filter in inline direction by rejecting all negative angles, then filtered result is again filter in cross line direction by keeping all positive and negative angles between 200ms/per km

Figure 15. Parameters for dip-filter in inline direction

Figure 16. Parameters for dip-filter in cross-line direction

Here we give one example of comparison before and after dip-filtering.

Figure 17. Comparison between unfiltered and dip-filtered section

From figure 17, we do see dip-filtering suppresses most of boundary reflection. But unfiltered section (middle image) looks more like real seismic section on the left. We can see some weak linear dipping events on seismic section too. Generally seismic is more noise comapring with modeled seismogram.

Time comparison

Figure 18. Extrcated 3 lines from modeled and real seismic

Figure 19. Extrcated 3 cross lines from modeled and real seismic

Now we display real seismic and modeled seismogram in mirror to see how time agrees each other. Figure 18 shows 3 inlines extracted from real seismic and modeled seismogram, modeled seismogram is mirrored, Figure 19 shows 3 extrcated cross lines with seismogram mirrored. We observed that events on modeled section showing lower dips than corresponding ones on real seismic section, this is because modeled seismogram is plane-wave simulation which is approximation of zero-offset post-stack data but real seismic data is migrated. Time difference is obvious for major events. Although push down modeled seismogram in time will make top of some events matching, but this is not a simply linear process. So only fix is fixing model itself.