A New Workflow for Surface Wave FWI Combining Genetic Algorithm and Gradient-Based Optimization Algorithms

Summary We propose a new workflow for surface wave (SW) inversion by combining the two-grid genetic algorithm (GA) and gradient-based full waveform inversion (FWI). The workflow circumvents the notorious requirement of having a “good enough” starting model” in gradient-based SW FWI. At the 1st step of the workflow, without any a-priori information, by employing a coarse inversion grid and by inverting the lower frequencies only of the observed data, GA SW FWI reconstructs the long-wavelength structures of the subsurface. Then, in the next step of the workflow, the GA predicted model is used as the initial model for gradient-based SW FWI. In virtue of the higher efficiency of the gradient-based method, finer inversion grids are adopted and data with higher frequencies can be inverted yielding refined predicted models. We discuss our approach making use of two synthetic examples that reproduce complex near-surface models and we show that fair inversion outcomes are obtained. Models predicted by GA SW FWI are proved to be adequate initial models for gradient-based SW FWI. In addition, the examples confirm the extremely strong impacts that initial models have on gradient-based SW FWI results.