Bridging spatiotemporal scales of normal fault growth during continental extension using high-resolution 3D numerical models: DATASET

This dataset supplements the publication <strong>“Bridging spatiotemporal scales of normal fault growth during continental extension using high-resolution 3D numerical models“</strong> by <em>Pan, Naliboff, Bell and Jackson</em> in <em>Geochemistry, Geophysics, Geosystems</em>. We include three 3D numerical simulations of continental extension developed using open-source, mantle convection and lithospheric dynamics code <em>ASPECT</em>, in addition to test models referenced in the manuscript, and python scripts to automatically extract the fault population. The purpose of the numerical modelling setup was to understand the growth of normal fault populations over 10⁴-10⁶ yrs. <br> Overall, the supporting files contain all the necessary parameter files and additional inputs required to reproduce the results of our publication. Here we provide a brief breakdown of what each file contains: <br> ‘Models’ contains the version of ASPECT we used to generate the models, and python scripts used to generate the initial compositional field and geothermal gradient - these inputs remain constant for each model. The ‘compositional field’ directory contains initial compositional field and the python script used to generate the input file, where plastic strain values of 1.5 and 0.5 are statistically distributed as blocks within the upper crust. Note that each run of the python script produces a new statistically random field. The model results are provided as interpolated numpy files, where <strong>Models 1, 2 and 3 correspond to extension rates of 2.5, 5 and 10 mm/yr, respectively</strong>. The specific ASPECT parameter file for each model can be found within each respective directory. Extracted fault data contains an excel database for each individual timestep, which we compiled together (named “Appended_relationships.xlsx”) for D-L fault evolution statistics. ’Final D-L statistics’ provide the compiled “Appended_relationships” for Models 1, 2 and 3, such that they are easily accessible for non-model specialists. <em>Each row corresponds to the geometric attributes of one fault, which are indexed.</em> The global compilation of D-L datasets, used to plot Figure 5, are also provided. ‘Fault extraction’ provides the script used to extract fault statistics. Two jupyter notebooks are provided to (1) process vtk data into an interpolated numpy file format, and visually showcase the image processing scripts used to extract fault labels, as well as showing the method to address the occurrence of bifurcated labels; and (2) reproduce the figures shown in the publication. ‘Movies’ which are referenced in the manuscript. The captions for all movies are provided here below. ‘Test models’ provide test model runs which supplement our results. Specifically, this folder contains test model runs which investigate the effect of:<br> <strong>Varying block sizes of initial binary plastic strain. </strong>1.25, 2.5, 5 and 10 km2 blocks were used, conducted at a lower 1250 m resolution. We found that a smaller block size corresponded to smaller faults overall. <strong>The inclusion of viscoplastic regularisation, in which a damper viscosity of 1e21 Pa s was used following methods from Duretz et al. 2020. </strong>We found that this damper produced similar fault patterns, but active deformation was relatively more diffuse (i.e. more blurred in appearance) and lower in strain magnitude. Note that the model with the plastic damper was run with a more recent version of ASPECT: commit cdb0333fa on main branch. <strong>The entire adaptively refined model extent, which reduced to its highest resolution (625 m) in the central portion.</strong> Due to size limits, we have included the last modelled timestep for Model 2. We found that the maximum fault lengths, which were only extracted from the centralised high-resolution portion, extend further into the lower-resolution zones. Although faults are expected to extend further, we also expect that faults are segregated due to their along-strike variability, but higher resolutions are needed to validate this observation. <strong>A stricter nonlinear tolerance of 1e-5 instead of 1e-4. </strong>We ran this test model up to 20 timesteps (or 2 Myrs extension duration). We find that a stricter tolerance produces a similar spatiotemporal trends and patterns strain rate second invariant field, however, the lowest magnitudes of active deformation (largely occurring at along-strike bends) are more detailed and resolvable. More prominent deformation occurs at the edges, which erroneously lead to anonymously high fault lengths in D-L plots. Significantly, we increased the maximum number of nonlinear iterations to 1000 when using a nonlinear solver tolerance of 1e-5, but only reached a residual of ~2.85574e-5 on the first time step. In subsequent time steps, the Stokes solver converged within 2-3 nonlinear iterations.