Dispersion based recurrent neural network model for methane monitoring in Albertan tailings ponds

Bitumen extraction for the production of synthetic crude oil in Canada's Athabasca Oil Sands industry has recently come under spotlight for being a significant source of greenhouse gas emissions. A major cause of concern is methane, a greenhouse gas produced by the anaerobic biodegradation of hydrocarbon in oil sands residues, or tailing, stored in settle basins commonly known as oil sands tailing ponds. In this work, we build a data-driven modeling framework to determine the methane emitting potential of these tailing ponds and have future methane projections using a Dispersion based Recurrent Neural Network (DIRNN). We show that our method can predict both methane emissions and concentrations by considering the transport of methane emissions in air, thereby outperforming existing other deep learning approaches. Using a reverse dispersion modeling approach, we use our trained model to identify active ponds and estimate about 56,303 tonnes of methane (1.5 million tonnes of carbon dioxide equivalent) emissions from the Athabasca oil sands tailings. Our results are consistent with previously reported emission estimates from various studies, and indicate atleast three times underestimation in official reports.