Predicting gas and hydrate presence on the U.S. Atlantic margin using geospatial machine learning

Abstract:

The occurrence of gas in shallow marine sediments, especially associated with methane hydrates, is important for understanding carbon cycling and submarine geohazards. Recent observations of seafloor methane venting near the updip limit of the gas hydrate stability zone on the U.S. mid-Atlantic margin suggest a relationship between hydrate dissociation and gas release, but this relationship has not been demonstrated in the subsurface. Scarce geophysical surveys make large-scale identification of gas difficult. We used a machine learning-based approach to predict the presence of gas in sediments on the shelf and continental slope. Our methods included using GPSM, a geospatial machine learning model, to predict total organic carbon (TOC) content at the seafloor on the mid-Atlantic margin, and using the TOC map along with other predicted seafloor properties as inputs to a 2-dimensional fluid flow model to predict gas generation through time and assess where hydrate and gas are likely to be located in the subsurface. The best prediction of seafloor TOC was obtained using a k-nearest neighbor algorithm. TOC abundance was predicted to be greatest (>2.5 wt%) along the upper continental slope between Hudson Canyon and Cape Hatteras, with the highest abundance in the vicinity of Norfolk Canyon and the headwall of the Currituck Slide. These correspond very well with the areas with the highest abundance of observed seafloor gas venting. Modeling indicates that these high TOC values drive microbial methanogenesis, resulting in subsurface accumulation of gas and hydrates. Seafloor gas venting in the region is likely directly tied to microbial methane abundance and could result either from gas accumulation or generation from dissociating hydrates.

SAND2020-7534 A