July 9 DGS Luncheon: Venkatesh Anantharamu “Seismic to Rock Property Mapping Using Rock Physics-Guided CNNs”

Date/Time
Date(s) - 07/09/2026
11:30 am - 1:00 pm

Location
Wynkoop Brewing Company

Categories

• Luncheon

July 9 DGS Luncheon: Venkatesh Anantharamu “Seismic to Rock Property Mapping Using Rock Physics-Guided CNNs”

Wynkoop, Thursday, July 9th

11:30-1:00

“Seismic to Rock Property Mapping Using Rock Physics-Guided CNNs”

Venkatesh Anantharamu, GeoSoftware

This study presents a machine learning–driven seismic quantitative interpretation workflow applied to an unconventional reservoir in the Midland Basin, demonstrating how reservoir and geomechanical properties can be directly predicted from pre-stack seismic gathers. A rock physics–guided convolutional neural network (CNN) architecture is developed to jointly estimate elastic properties, reservoir properties, and mineral composition, addressing the challenge of resolving fine-scale heterogeneity within highly laminated mudstones, siltstones, and thin carbonate units.

The approach integrates well logs, synthetic AVO modeling, and seismic gathers to build a large and diverse training dataset that captures realistic variations in lithology, fluid content, and mechanical behavior. Forward modeling using the Keys and Xu rock physics framework enables the generation of hundreds of “what-if” synthetic scenarios, providing the variability needed for robust CNN training. Two CNN models are trained: the first predicts porosity, water saturation, P-impedance, S-impedance, and density, which are then used to compute geomechanical properties such as Young’s modulus and brittleness index. The second CNN model simultaneously estimates mineral volumes including quartz, clay, calcite, and kerogen. Transfer learning further refines the models by incorporating real seismic gathers and well-log responses, reducing synthetic–to–field mismatch and improving prediction stability.

Model outputs show strong agreement with measured logs across the study area, including six blind validation wells. The predicted density, porosity, and brittleness index consistently resolve the target interval and exhibit coherent lateral continuity across the section. This level of quantitative accuracy highlights the advantage of rock physics–guided CNNs in producing physically meaningful, well-calibrated predictions. Overall, the workflow enhances reservoir characterization, improves understanding of mechanical stratigraphy, and provides a practical tool for identifying brittle, hydraulically favorable zones that support drilling and completion optimization in the Midland Basin.

Speaker Bio

Venkatesh is a geophysicist with over 12 years of experience in seismic interpretation, quantitative reservoir characterization, and rock physics–driven workflows across diverse onshore and offshore basins. His work integrates advanced geophysical methods with machine learning techniques to improve reservoir property prediction and reduce subsurface uncertainty.

At GeoSoftware, Venkatesh collaborates closely with product, research, and services teams to develop and apply innovative technologies that enhance reservoir understanding and drive practical business outcomes.

If the PayPal option does not allow you to pay with credit card, you can pay with a card in person at the luncheon.  Please use the Pay by Check option at checkout so we can get a good headcount.

Bookings