Machine-learning algorithms have gained acceptance in the industry to estimate petrophysical properties and production metrics (Ross, 2019). Reservoir models using the estimated properties, and incorporating Discrete Fracture Network models, are another practice for better understanding and predicting unconventional reservoir behavior. However, these new technologies are rooted in proper execution of basic seismic-interpretation practices. These include determining the suitability of acquisition geometries, the implementation of appropriate imaging/inversion algorithms, and careful formation of the interpretive database. Database issues include proper loading and QC of well information such as headers, deviations, logs, and well tops. Similar consideration must be given to the seismic data.
The points above are best evaluated in context of employing deliverables obtained from anisotropic depth imaging. The higher fidelity of the data allows for detailed synthetic ties of well tops to seismic inversion volumes to establish horizon correlations. A joint review of well tops with their equivalent seismic depth interpretations then facilitates the identification of outliers in the database. This is a critical process, as inconsistencies not detected and resolved in the database will likely yield misleading results that can compromise a company’s multidisciplinary-team efforts.
Having established a consistent database, an example is given of estimating lateral-length normalized EUR using a machine-learning algorithm and seismic attributes derived from prestack inversion.