Quantifying CH4 and NO2 Emissions Using Satellite Measurements with Environmental Impact in Texas: A County-Wide Assessment for Oil & Gas Industry

Abstract

Abstract This study presents a comprehensive machine learning and data science approach to predict and quantify ground-level methane concentrations using high-resolution satellite data, climate data, and production data. Our study focuses on quantifying ground-level methane concentrations in the Permian Basin, within Texas state, from January 2022 to December 2022. The Machine Learning (ML) pipeline includes data pre-processing, feature selection, model training, and evaluation. The dataset comprises TROPOMI methane mixing ratios and nitrogen dioxide columns, ERA5 climate variables, ground-level methane measurements from towers, and oil and gas production data. This study leverages Random Forest, Gradient Boosting (Light and Extreme) models, with hyperparameter tuning to optimize their performance. The results indicate that Extreme Gradient Boost model outperforms the other models, exhibiting a superior fit with minimal errors and accuracy of 95%. We also conducted feature importance analysis that highlights the significant contributions of nitrogen dioxide, air temperature, and methane mixing ratios with gas well gas production for predicting ground-level measurements. Additionally, the study demonstrates the spatial variations of methane concentrations county-wise on a weekly basis, with Winkler County exhibiting notably higher levels compared to other counties, providing valuable insights for environmental decision-making. These findings contribute to informed decision-making and targeted mitigation strategies for reducing methane emissions in the region.