Efficient Predictions of Global Free Gas and Gas Hydrate Formation using K-means Clustering
Abstract
Significant quantities of free gas and gas hydrate contained beneath the seafloor are crucial for climate modeling, carbon budget estimates, and determining acoustic velocity of seafloor sediment. While free gas and gas hydrate have been discovered at various global locations, the availability of geophysical data required for accurate predictions of their occurrence is sufficient in areas of active oil or gas production but remains scarce or absent in critical regions such as the Arctic. To compensate for the variations in data coverage, researchers at the Naval Research Laboratory used machine learning techniques to extend geophysical information from previously studied regions to poorly constrained areas to produce the Global Predictive Seabed Model (GPSM). We have developed a workflow that couples Dakota to PFLOTRAN to probabilistically predict free gas and gas hydrate occurrence. Dakota uses Latin hypercube sampling of the GPSM values and their uncertainties to determine distributions which are used as PFLOTRAN input parameters to simulate methanogenesis and predict hydrate and gas formation. We apply k-means clustering to the GPSM data from a study area of ~24,000 offshore locations between Svalbard and Norway (10°E - 30°E, 70°N - 80°N) to determine a subset of simplified clusters characterized by similarities in sedimentation rate, TOC, heat flux, temperature, and depth. Every region is described by a set of means and standard deviations for these parameters that are sampled on by Dakota to generate input decks for PFLOTRAN simulations. We ran 500 simulations for each cluster and map the probabilities of free gas and gas hydrate formation to their corresponding geographic regions. To verify the process, we also ran 50 simulations at all offshore locations in the study area and find strong agreement for free gas (r=0.967) and gas hydrate (r=0.947) formation rates from both the k-means and individual simulations. Both simulation methods predict elevated formation rates of free gas and gas hydrate in shallow regions between Svalbard and Norway. The k-means technique was then extended to the full GPSM dataset to make probabilistic predictions of global occurrence. This efficient technique provides preliminary predictions that identify important regions of gas and hydrate accumulation in seafloor sediment.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFMOS25A1003E