Himalayan Geology, Vol. 45 (1), 2024, pp. 89-107, Printed in India

Machine learning assisted gas hydrate saturation proxy: A case study from KG basin, India

SUMAN KONAR¹, BAPPA MUKHERJEE², KALACHAND SAIN²*

¹Control Source Seismics and Gas Hydrates, CSIR-National Geophysical Research Institute, Hyderabad-500007, India

²Seismic Interpretation Laboratory, Wadia Institute of Himalayan Geology, Dehradun-248001, India

*Email (Corresponding author): kalachandsain7@gmail.com

Abstract: The study proposed implementation of machine learning algorithms for estimating gas hydrate saturation using geophysical logs. We used three supervised machine learning techniques: Decision Tree, Extreme Gradient Boosting and Gaussian Process Regression. The study demonstrates the practicality of the proposed method with well log data from the Krishna-Godavari offshore basin in India, addressing limitations of traditional methods such as resistivity and acoustic log based method. Initially, modified Archie's and Indonesian equations were used to estimate gas hydrate saturation (S_h) from geophysical logs at two wells in a clay-dominated region. The ML models were trained using these log data as inputs and corresponding traditionally estimated saturations as output. Further, the gas hydrate saturation was predicted at two neighbour wells using their downhole logs by utilising the pre-existing data driven pattern among the logs and S_h. The model accuracy was measured through root mean square error and correlation coefficient calculation during training and validation stage. Present study reveals that Decision Tree yields the best prediction performance in computing S_h among the three implemented methods. The predicted saturation in the studied sites varies between 0% to ~43%. The demonstrated method provides very precise values of gas hydrate saturation which is an essential parameter required for accurate resource evaluation and well-planned production.

Keywords: Gas hydrate, Machine Learning, KG basin