The data set includes the 3D incremental displacement fields resulting from Digital Image Correlation (DIC) analysis of four strike-slip experiments performed at the laboratory for experimental tectonics at GFZ Helmholtz Centre for Geosciences in Potsdam in 2022. The data here include the incremental displacement time series from four strike-slip experiments with quartz sand G12 (Rosenau et al., 2018): two with 1.2 cm thick sandpack and two with a 3 cm thick sandpack. Ramos Sánchez et al. (in review) used the incremental horizontal displacement fields from strike-slip fault experiments within different materials to train Convolutional Neural Networks (CNN) to predict off-fault deformation from active fault trace maps. Because off-fault deformation is difficult to ascertain in the field, convolutional neural networks trained on data from scaled physical experiment that simulate upper crustal deformation can inform how much off-fault deformation can be expected along crustal strike-slip faults.
For training the CNNs Ramos Sánchez et al. (in review) used incremental horizontal displacement fields from experiments in wet kaolin (Cooke et al., 2021) and both poured and sedimented CV32 sand (Visage et al., 2023). All experiments used identical conditions of straight basal velocity discontinuity to produce overlying strike-slip faults. The benefit of training the CNN on strike-slip experiments within different materials is to capture a wide range of strike-slip deformation that may occur within the upper crust. After training of the CNN Ramos Sánchez et al. (in review) tested the trained on unseen fault maps including maps from the two 3 cm thick G12 sand experiments of this dataset. The 1.2 cm sandpack maps were excluded from the study because the strike-slip faults were very fine and closely spaced so they were not as well resolved as those of 3 cm sandpack. By testing the CNN trained on CV32 fault maps with the fault maps from G12 sand experiments, Ramos Sanchez et al. were able to assess if the applicability of the CNN to experiments with similar but not identical sand.
Information on the displacement field analysis and CNN training and testing can be found in the main text and supplement to Ramos Sánchez et al. (in review). The file structure of this zip folder is fully described in the list of files.
The data set includes the digital image correlation of 16 dextral strike-slip experiments performed at the University of Massachusetts at Amherst (USA). The DIC data sets were used for a machine learning project to build a CNN that can predict off-fault deformation from active fault trace maps. The experimental set up and methods are described with the main text and supplement to Chaipornkaew et al (in prep). To map active fault geometry and calculate the off-fault deformation we use the Digital Image Correlation (DIC) technique of Particle Image Velocimetry (PIV) to produce incremental horizontal displacement maps. Strain maps of the entire region of interest can be calculated from the displacements maps to determine the fault maps and estimate off-fault strain throughout the Region of Interest (ROI). We subdivide each ROI into five subdomains, windows, for training the CNN. This allows a larger dataset from the experimental results. The data posted here include the incremental displacement time series and animations of strain for the entire ROI.
Experiments of oblique convergence at angles of 5, 10, 15, 20, 25 and 30 degrees from the margin within wet kaolin. One suite of experiments, denoted as ‘precut’, has a vertical surface precut within the clay with an electrified wire. The precut surface lies directly above the basal oblique dislocation. The other suite of experiments is ‘uncut’. Regardless of whether the experiments have a precut surface, slip partitioned fault systems, develop and persist in the experiments. Such systems have two simultaneously active faults with similar strike but different slip sense. Slip partitioning also develops regardless of whether the system first grows a reverse fault or strike slip fault in the experiment. The sequence and nature of strike-slip and reverse fault development depends on present of existing cut and convergence angle.
This data set includes time series of incremental displacement maps for eleven experiments performed at the University of Massachusetts Amherst in January 2017 and March 2018 as well as animations of strain and uplift. The dataset includes the 30˚ convergence experiment with precut vertical surface but the 30˚ uncut experiment has not yet been performed. The time series data are organized into 11 netCDF files. The name of each file states the obliquity of convergence and whether the vertical surface was precut or not.
Each netCDF file contains the following
• ux = the incremental displacement field within the ROI (Region Of Interest) parallel to the margin (x-direction). The third dimension in the array corresponds to increment of deformation through the experiment. Units are mm.
• uy = the incremental displacement field within the ROI perpendicular to the margin (y-direction). The third dimension in the array corresponds to increment of deformation through the experiment. Units are mm.
• x = position parallel to the margin. Units are mm.
• y = position perpendicular to the margin. Units are mm.
The incremental displacements are calculated from DIC of photographs taken every 30 seconds using PIVlab (Thielicke, 2019). The net stepper motor speed is ~0.5 mm/min.
The animations show strain evolution of all eleven experiments and uplift evolution of the 10 degree precut experiment. The strain evolution experiments overlay colormaps of incremental strain between successive photos on photographs of the experiment. Color saturation indicates the strain rate and hue indicates the slip vector. The uplift maps were made from stereovision analysis from pairs of photos. In most experiments, decorrelation of portions of the map prevented us from producing high quality uplift evolution animations from the start to the end of the experiment. Only the 10 degree convergence with precut vertical surface experiment had full coherence of uplift signal throughout the experiment and that animation.