Reflection seismic has limited bandwidth by nature. It starts with the lower part of the spectra in 8 to 10 Hz for traditional recording (though with actual recording geophones and sources it can be as low as 1 or 2 Hz) and in the upper end it reaches up to 65 to 80 Hz (again depending on many factors). Encouraged by the important amount of available well control points we propose to recover a broader band spectrum seismic (HF) than in the original by two methods based on Seismic Inversion as Impedance input:
a. Convolution of vertical reflectance coefficients log in time with a wavelet of desired bandwidth. Reflectance was previously derived from a strong resolution enhanced P-Impedance using neural networks (CNN).
b. Direct neural network prediction with synthetic seismic as target (MAR and DNN).
Throughout this work several tests are mentioned: comparison of original seismic with high frequency seismic, synthetic seismograms for both, comparative geometric attributes, differences in interpretation values both in time and amplitudes. All of them applied to an area of the unconventional shale-oil of Vaca Muerta Fm, Neuquina Basin, Argentina.
The search for higher seismic vertical resolution is intended to achieve a more effective link with the reservoir geology for different areas of hydrocarbon production. Thus, many methods and algorithms are generally tested so that the bandwidth could better represents the geology present and lateral variations.
In this case we start from a basic premise related to the inversion of seismic traces: the wavelet is obtained in a bounded time window where it is accepted as stationary, i.e., it does not undergo changes in its definition characteristics. In that case, a result of higher frequencies will be obtained from the inversion, in the form of Acoustic Impedance. Besides, there is calculated the Full Bandwith or Full Frequency Band, where the low frequencies are part of the inversion coming from the filtered low band pass average impedance of the profile of the well log.
