Assessing model uncertainties for joint inversions of seismological data using a genetic algorithm

Error bars were generated for velocity models using receiver functions and surface wave dispersion curves for four seismic stations in southern Africa, with a genetic algorithm adapted from the code NSGA-II. Each receiver function and dispersion curve was originally created by Eldridge Kgaswane (2009). We examined these stations, and through a series of statistical resampling, we were able to place an uncertainty on each layer's velocity in the lithosphere. Each station was set to an initial model, which was perturbed to generate a series of best-fit models for the corresponding receiver functions and dispersion curves. For each layer of depth, a series of solutions evolve over a set number of generations using "survival of the fittest" to come up with these best-fit models. These were constrained to only consider geologically viable models, such as the velocity range in each layer and smoothing. Afterward, the error bounds on velocities were able to be placed on each layer. The velocity vs. depth plot gives the uncertainty from 1 to 2.5km in depth. Now a better estimate of the velocities of the waves can be made, which leads to a better estimate of the composition of the lithosphere under southern Africa.