AUTHOR=Suleymanov Vagif , El-Husseiny Ammar , Glatz Guenther , Dvorkin Jack 

TITLE=Rock physics and machine learning comparison: elastic properties prediction and scale dependency

JOURNAL=Frontiers in Earth Science

VOLUME=Volume 11 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/earth-science/articles/10.3389/feart.2023.1095252

DOI=10.3389/feart.2023.1095252

ISSN=2296-6463

ABSTRACT=Rock physics diagnostics (RPD) established based upon the well data are used to deterministically predict elastic properties of porous rocks from measured petrophysical rock parameters. With the recent advances in statistical methods, another approach for predicting elastic properties from well data is through the use of machine learning (ML). Several studies have reported the comparison of rock physics and ML methods for the prediction of elastic properties. However, the scale independence (applicability of results at both well log and seismic scale) of the ML approach was never investigated. This study aims at comparing the results from rock physics and ML methods for predicting elastic properties, namely the bulk density (ρb), P-wave velocity (Vp), S-wave velocity (Vs), and Poisson’s ratio (PR) in a well from the Gulf of Mexico (GOM). In particular, well logs, such as porosity, gamma ray, and resistivity curves were used as ML inputs to simulate the petrophysical inputs employed in rock physics models. The well-log data were upscaled, using Backus averaging, to examine the scale independence and prediction accuracy of physics-driven and data-driven approaches at the seismic scale. The evaluation of predicted elastic properties was based on the average absolute percentage error (AAPE) between measured and predicted outputs. Results show that ML models provide better prediction accuracy at the well log and seismic scales compared with the RPD. On average, RPD has an AAPE of 4% for Vp prediction and 10% for Vs and PR prediction while the AAPE for ML does not exceed 3% for all elastic properties. Both RPD and ML provide consistent results at both well log and seismic scale, suggesting the scale independency of both approaches. Nevertheless, the ML models failed to identify an issue in one gas-saturated interval showing non-realistic Vs and PR values. Such an issue, however, could be identified and corrected using the RPD, which shows the importance of domain knowledge to check data quality and validate results. This study suggests the importance of incorporating rock physics modeling even when using machine learning methods in modeling elastic properties to check the quality of the training data and ensure reasonable predictions.