Abstract

Methane emissions are a crucial environmental concern for climate change. Therefore, locating, quantifying, and mitigating these emissions sources is essential. Methane signatures from orphan wells can be detecting from drones equipped with the appropriate sensors but a methodology is required to estimate methane flux from the measurements made by the drone. We trained a convolutional neural network (CNN) using Large Eddy Simulations (LES) dataset of methane plumes that mimic the real dataset of the next-generation Airborne Visible/Infrared Imaging Spectrometer (AVIRIS-NG) where wind speeds are varied from 1 m/s to 4 m/s with varying source flux rates of 20-100 kg/hr. We adopted a deep neural network architecture, similar to Visual Geometry Group (VGG), to train our model on synthetic data efficiently. The model architecture is much simpler than VGG and is very fast to train. We optimized the model parameters and architecture to improve the model’s precision and quantified the source flux rates of methane plumes with high accuracy and tested our trained model using synthetic 2D images. In the synthetic dataset, the overall predictability of source flux rates from methane plume data is excellent, and the mean absolute percentage error for predicting the true source flux rate values of all the training images is 4.31%, for all validation images is 12.4%, and for all test images is 8.27%. Once validated against field data, a machine learning model such as this can be used as a screening tool to determine which orphan wells are leaking large amounts of methane so that these wells can be prioritized for plugging and abandonment.

Introduction

Methane (CH4) is the second largest greenhouse gas (EPA, 2022) after carbon dioxide (CO2) in the Earth’s climate system and one of the most important anthropogenic greenhouse gases. Methane is flammable and can pose safety hazards in places like oil and gas facilities, orphan and abandoned wells, house basements, and closed places. Moreover, CH4 ultimately produces CO2; tropospheric ozone, and stratospheric water vapor after different reactions with the atmospheric gases. They all contribute to climate change and indirect radiative forcing in the long run. However, methane has a shorter-state lifetime (Prather et al., 2012) than most other greenhouse gases (e.g., CO2), and its loss is primarily the result of atmospheric oxidation (Montzka et al., 2011). Different studies indicate that CH4 emissions using atmospheric measurements are higher than those reported from inventories (Jeong et al., 2013, 2014; Wong et al., 2016), and CH4 emissions from the oil and gas supply chain are about 60% higher than those reported in the national greenhouse-gas inventory (Zavala-Araiza et al., 2015). Both anthropogenic and natural emissions of CH4 are likely to increase if no mitigation plans are deployed.

This content is only available via PDF.
You can access this article if you purchase or spend a download.