Prediction of Reservoir Quality Sands Using Seismic Attributes and Geomorphology Driven By Machine Learning Facies Classification In Lower Indus Basin, Pakistan

Abstract

Assessing reservoir quality plays an instrumental role in an oil/gas fields performance and lifecycle. Lower Goru gas fairway is a prolific producer in the Lower Indus Basin. This research is an effort to demarcate the reservoir quality within the different system tracts and parasequence sets. High energy depositional environments have better reservoir quality as a general rule, and we apply this concept to evaluate the East-West oriented progradation, parasequence stacks and system tracts. The hydrocarbon-bearing intervals of Miano gas field belong to the Early Cretaceous and are bounded by multiple flooding surfaces and other unconformities identified through well correlation. Overall, the NTG is good and porosity ranges from 12-16% in zones of economic water saturation (<30%). Seismic interpretation based on local flatness and semblance attributes highlights progradation and a basinward shift of facies. The upper shoreface facies deposited by the wave-tide effect has better amplitudes and reflector continuity. The reservoir quality gets affected in lower sequences as the shale and heavy mineral content increase. Attributes including Grey Level Co-Variance Matrix (GLCM), 3D Curvature and Variance are rendered on the seismic volume for improved feature detection and holistic evaluation. These attributes are then blended for improved constraint on the deposition-facies association. In order to better quantify reservoir quality sands in the area, we run machine learning algorithms including Principal Component Analysis (PCA) and Self Organizing Maps (SOMs) on seismic attribute data and (MLPSVM) Multilayer Perception Support Vector Machines model to predict facies using RGB logs with accuracy of 92.31%. The results add immense value by sharply amplifying subtle stratigraphic details and predicting reservoir quality across a varying stratigraphic ecosystem. To better understand the reservoir quality variation within our area of interest, seismic data is inverted to impedance and clay volume estimated from logs along with porosity is populated on the impedance volume. The sand is distributed as lenses with shale deposited in between them. With good porosities and low clay volume, these sand lenses provide good locations for future exploration endeavors. Furthermore, river ravinement, channels and lobe trends can also be observed on the inverted volume. The effective porosities are high, and clay volume is low in sand bodies in the upper inverted sequence, reflecting good reservoir quality. Evaluating log signatures for stratigraphic control over facies and seismic interpretation providing insights into the facies variation trend at different levels, the machine learning steered attribute analysis and inversion at the upper shoreface sequence above the flooding surfaces agree with each other thus encouraging the validity of results at the lower transgressive and regressive sequences in Miano.i