GRSG 35th Conference 2024 Abstract

Title: Evaluating Machine Learning Algorithms and satellite-based covariates to spatially detect active petroleum seepages

Author: Mohammad Hassan Tayebi

Organisation: National Karst and Hard Rock Research center, Iran

Evaluating Machine Learning Algorithms and satellite-based covariates to spatially detect active petroleum seepages

Mohammad H. Tayebi, Diego Fernando Ducart, Carlos Roberto de Souza Filho, Saeid Asadzadeh, and Majid H. Tangestani

The Zagros Fold-and-Thrust Belt (ZFTB) of Iran is one of the world’s largest petroleum producing basins, with numerous active and inactive natural petroleum seepages. This study applies and evaluates Machine Learning (ML) algorithms comprising Random Forest (RF) and Support Vector Machine (SVM) integrated with satellite-based covariates to detect active petroleum seepages in the Aghajari oil field, SW Iran.

The spectral and geochemical properties of gelogical formations and the Koh-e-Sokhteh active petroleum seepage in the Gachsaran Formation as the main evaporite cap rock were measured, characterized, and compared to the un-altered rock units. The Boruta Algorithm, Pearson’s correlation coefficient and variable importance plots were used to identify the most important satellite-based covariates, assess the strength of the covariate linear relationship, and determine the importance of covariates, respectively. The ML models were trained by randomly splitting field samples into two groups (70% for training and 30% for model testing), and their accuracy was measured using the Overall Accuracy (OA) and Kappa Coefficient (KC) metrics. Spectroscopic analysis revealed meaningful changes in the spectral properties of evaporitic rocks beween altered and unaltered units. The RF algorithm performed slightly better than the SVM algorithm, achieving an OA of 95% and a KC of 0.89, compared to the SVM’s OA of 90% and KC of 0.79. Active petroleum seepages were mapped in the southeastern flank of the Aghajari oil field. The findings of this study can be applied to other basins with similar geological settings, providing a valuable tool for the petroleum seepage detecting and environmental monitoring.