Haroon, Amir , Paasche, H., Jegen, Marion , Graber, Sebastian and Petersen, Sven (2021) Using machine learning workflows to identify seafloor massive sulfides based on disparate geophysical data. [Poster] In: 81. Jahrestagung der Deutschen Geophysikalischen Gesellschaft (DGG). , 01.03.-05.03.2021, Kiel (online) .

Abstract

Many current research questions in Earth Sciences are related to understanding the complex geological processes that dictatethe location of resource formation. Multivariate and multi-disciplinary measurements across disparate spatial scales are common and generate databases that cross traditional geoscientific domains such as geophysics, geochemistry and geology. To cope with this mass of information, integrative data assessment approaches are essential for optimal information extraction from the available data, enabling more accurate prognoses of where to find previously undiscovered natural resources. This is especially true in the deep ocean, where data collection is inherently more difficult than on land. Here, we present an example from predicting the spatial distribution of seafloor massive sulfides (SMS) at the Trans-Atlantic Geotraverse (TAG) hydrothermal field using various sources of marine geophysical data, including high-resolution bathymetry and magnetics collected with an autonomous underwater vehicle (AUV), as well as conductivity data derived from Controlled-Source Electromagnetic (CSEM) measurements. Electrical conductivities are considered a direct SMS indicator as these exhibit a substantial conductivity variation to the surrounding host rock. Due to the size of the survey area, a spatial sampling of conductivity on a dense grid is not economical and would require many years of data acquisition. As a result, robust extrapolation of sparsely sampled conductivity data onto a regional scale seems efficient for predicting additional occurrences of SMS by integrating the acquired marine geophysical data into a Data Science framework. We show that this framework may improve current SMS predictions through associated prediction uncertainty, thereby detecting observational gaps and informing further sparse sampling of ground-truthing data (e.g. sampling or visual observations).

Actions (login required)

View Item