Strain evolution in continental rifts: Machine learning of seismic reflection data and numerical models

• Applicant: Professor Dr. Sascha Brune
• Subject Area: Geophysics ; Geology
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 460760884

Continental rifting is a key geodynamic process that continuously shapes the surface of our planet. Presently active rifts and continental rifted margins host significant natural resources, such as groundwater and geothermal energy, but are also sites of strong, infrequent earthquakes posing considerable natural hazard. To understand the occurrence of geohazards and georessources, we need to quantify how strain localises and how fault networks evolve in crustal-scale extensional systems. To this aim we propose an innovative, multidisciplinary project that develops machine learning techniques in order to analyse big data sets from both 3-D seismic reflection surveys and 3-D numerical geodynamic models. This will effectively bridge a major gap between these two branches of solid Earth research. Our machine learning approach will put us in a unique position to automatically investigate large volumes of data, which are impossible to analyse manually, shedding new light on three fundamental sets of question regarding the dynamics of extensional fault networks in space and time:Q1) How is rift strain distributed spatially? How much does it vary with depth?Q2) How does the 3-D strain field evolve during fault network evolution? What is the impact of crustal rheology?Q3) How does strain vary spatially if the orientation of extension changes? Can we use fault orientations to infer changes in plate motions?Here we apply for a three-year research project, in which we will use deep convolutional neural networks to extract fault displacement from 3-D seismic reflection data. Our focus region will be the northern North Sea rift, where we will analyse a unique new seismic dataset that covers almost the entire rift (35,410 km2) at a resolution of a few tens to hundreds of meters imaging into a depth of ~20 km. The observational fault network analysis will be complemented by geodynamic models, which we as well analyse through machine learning tools. In contrast to the seismic data analysis, the numerical modelling approach will simulate the full time-dependent evolution of a rift system offering the possibility of a detailed spatio-temporal strain analysis. Ultimately, we will conduct and analyse a model suite for a range of representative rheological and kinematic parameters in order to infer the variability in deduced strain patterns and to quantify the large-scale extensional history of the northern North Sea. Our results will elucidate extensional fault network dynamics that are of high relevance to many other rifts and rifted margins worldwide. Machine learning techniques hold an enormous potential for analysing big data sets in similar and other study areas. We will therefore share all our tools developed during this project under an open source license.

• DFG Programme: Research Grants
• International Connection: Norway, United Kingdom
• Cooperation Partners: Dr. Rebecca Bell; Professor Dr. Robert Gawthorpe; Dr. Indranil Pan