BRINGING MARINE GEOSCIENCE INTO THE BIG DATA ERA: A NECESSARY APPROACH WHEN ‘I DON’T KNOW’ ISN’T GOOD ENOUGH

Marine geoscience has historically been advanced by studies utilizing various geophysical and direct sampling methods to make detailed characterizations of relatively small (~1000 km² scale) areas of the seabed. This approach has greatly advanced the marine geoscience state of the art, but is increasingly suboptimal due to increasing expense and administrative complexity of open marine field efforts. While autonomous platforms such as gliders and Saildrones offer great promise for reducing the expense of new marine geoscience data acquisition, there also exists a vast trove of legacy data. These legacy data can be utilized to characterize areas of the seabed that may no longer be logistically or geopolitically inaccessible.

Here we present a sampling of machine learning efforts undertaken by our research group over the last five years, largely utilizing data mining and geospatial machine learning. These efforts provide first order constraints on a variety of important marine geoscience parameters, such as heat flow, isopach thickness, and sediment accumulation rate. The geospatial predictions also identify areas where new data would most improve predictive skill, maximizing the utility of expensive field efforts. We view data driven marine geoscience as a necessary innovation to best utilize all existing data, and to quantify important marine geoscience parameters where traditional data acquisition techniques are not possible.