3D Modelling of Seismically Active Parts of Underground Faults via Deep Learning

Abstract

<div> <p>Underground faults are strain energy storage elements, the partial release of which is being depicted as a series of earthquakes, in the form of foreshocks-main seismic event-aftershocks, ones it reaches the surface of the Earth’s outer crust. Several underground faults, mostly bellow mainland, have been well mapped in terms of their three-dimensional extent. This is hardly the case though for underground faults beneath the sea. In the case of the latter, expensive efforts have produced two-dimensional mappings mostly in areas of hydrocarbon interest, such as the area of the Southern Hellenic Seismic Arc in the Eastern Mediterranean Sea. The challenge lies in deriving the in-depth extent of the latter when the surface of the Earth is covered by up to 3-4km of sea water as is the case in parts of the Eastern Mediterranean Sea. To that respect, known earthquakes’ hypocentres can provide valuable information regarding the in-depth extent of the seismically active part of the particular underground fault that caused the aforementioned earthquakes. For this to be achieved though it is of impeccable importance to correctly associate a seismic sequence not just to a particular underground fault, but to the certain part of it that gave rise to the observed earthquakes. This task falls well within the capabilities of learning neural networks, which can be trained with known pairs of earthquakes and associated underground faults to map one to another. Several difficulties arise in terms of overtraining due to the small training data set available and due to the fact that most of the training data comprise of mainland underground faults rather than underground faults in the crust beneath the sea. Still, in the cases were results from the deep learning neural network are deemed to be successful, the assertion of the vertical component to the initial two-dimensional mapping of underground faults becomes possible, producing three dimensional models of the latter. Finally, CUDA heterogeneous parallel processing enables 3D imaging and navigation amongst the spatial dataset of the underground faults situation in the Earth’s crust.</p> <p>Indexing terms: underground faults mapping, deep learning, modelling, 3D imaging, CUDA</p> <p>References</p> <p>Axaridou A., I. Chrysakis, C. Georgis, M. Theodoridou, M. Doerr, A. Konstantaras, E. Maravelakis. 3D-SYSTEK: Recording and exploiting the production workflow of 3D-models in cultural heritage. IISA 2014 - 5th International Conference on Information, Intelligence, Systemsand Applications, 51-56, 2014.</p> </div> <p>Konstantaras A. Deep learning and parallel processing spatio-temporal clustering unveil new Ionian distinct seismic zone. Informatics. 7 (4), 39, 2020.</p> <p>Konstantaras A.J. Expert knowledge-based algorithm for the dynamic discrimination of interactive natural clusters. Earth Science Informatics. 9 (1), 95-100, 2016.</p> <p>Konstantaras A., M.R. Varley, F. Vallianatos, G. Collins, P. Holifield. Recognition of electric earthquake precursors using neuro-fuzzy methods: methodology and simulation results. Proc. IASTED Int. Conf. Signal Processing, Pattern Recognition and Applications (SPPRA 2002), Crete, Greece, 303-308, 2002.</p> <p>Moshou A., P. Argyrakis, A. Konstantaras, A.C. Daverona, N.C. Sagias. Characteristics of Recent Aftershocks Sequences (2014, 2015, 2018) Derived from New Seismological and Geodetic Data on the Ionian Islands, Greece. Data. 6 (2), 2021.</p>