Date of Award
Doctor of Philosophy (PhD)
Kevin H. Mahan
Preserved lower-crustal terranes provide a view into deep earth processes unparalleled in spatial and temporal resolution. We use this rich data set of field observations and microstructural samples to document long-term strain localization in the dry lower crustal root of a continental strike-slip fault system, model the passive seismic properties its deformation, and create a novel method of interpolating missing orientation data using neural networks.
The Cora Lake shear zone is a 4-6 kilometer wide and >90 km long zone of mylonite and ultramylonite which hosts frictional melt (pseudotachylyte) veins with various intensities of ductile fabric overprint. High-resolution electron microprobe analyses constrain rupture to deep crustal conditions contemporaneous with all stages of ductile deformation in the shear zone. The cyclic rupture and ductile overprint of pseudotachylyte in the Cora Lake shear zone suggests a strong lower crust with a complex connection between ductile root and overlying brittle fault system. Experimental flow laws combined with microstructural observations predict a major viscosity contrast across the shear zone, resulting in stress amplification and the formation of the ~1km wide zone of ultramylonite intermixed with pseudotachylyte. These high stresses likely set the stage for deep aftershocks triggered by slip on the shear zone’s upper crustal extension.
In order to determine whether an analogous active or relict structure would be visible to geophysical techniques in the subsurface setting today, we model the passive effect of deformation-induced crystallographic preferred orientations and mylonitic layering in the Cora Lake shear zone. Although the mylonites in the Cora Lake shear zone produce low Vp and Vs anisotropy due to destructive interference between deforming minerals, mylonitic and gneissic layering contributes significantly to the magnitude and geometry of anisotropy.
Lastly, we use ~400,000 crystallographic orientations collected from the Cora Lake shear zone to test the effectiveness of a machine learning technique in predicting missing values against the common neighbor-averaging approach. Our neural networks outperform neighbor-averaging, and do so by apparently exploiting previously unknown predictive relationships within technique-specific orientation pattern metadata. This illustrates the potential value of machine learning approaches in the ever more data-driven earth sciences.
Orlandini, Omero Felipe, "The Cora Lake Shear Zone: a Natural Laboratory in the Continental Lower Crust" (2019). Geological Sciences Graduate Theses & Dissertations. 156.