3D Probabilistic Full Waveform Inversion: Application to Gulf of Mexico Field Data

Abstract

Summary Creating accurate models that represent spatial variations in subsurface velocity is a key component of seismic processing and imaging workflows. Over the last 10 years a significant tool used in this process is Full Waveform Inversion (FWI) which matches modelled to observed seismic data by inverting for velocity model estimates. In most cases, when FWI is applied, the focus of its application is on creating the single most accurate (“best”) model possible. In this work an alternative approach is taken, using Stein Variational Gradient Descent (SVGD) optimisation together with FWI to estimate subsurface velocity and the uncertainty in the estimate. To demonstrate this method, we use 3D seismic data collected by ocean bottom nodes (OBN) from the Gulf of Mexico. Through this practical field-scale example we demonstrate that SVGD FWI can provide a set of models with multiple plausible values for subsurface velocity. The subsurface locations where these models converge to similar values (i.e., the most certain) correspond with areas well sampled by the seismic data. This method is not without limitations and the computational cost is high, but the results presented are encouraging, with a range of potential applications that could help reduce risks associated with velocity model uncertainty.