Towards high-resolution local earthquake body-wave imaging using DAS in areas with exceptionally low seismicity rates

Detailed images for determining fine-scale geological structure are often best achieved through seismic reflections, which are sensitive to velocity gradients. Through application in a ‘big data’ framework with closely spaced receivers and shots, the hydrocarbons industry has demonstrated just how effective this approach can be. However, both the logistical effort and expense of this implementation are prohibitive for most applications. Recently, DAS is democratising the realm of big data, at least on the receiver side. The question we ask is: can we use DAS receiver ‘big data’ for high-resolution imaging, even in the absence of ‘big data’ on the source side?

Here, we propose a way forward using Fourier Neural Operators (FNOs), which are powerful at mapping between functions (e.g. a seismic wavefield and velocity models). As a proof of concept in the numerical domain, we use FNOs to invert for 2D P-wave velocity models from single earthquake gathers.

We first create a dataset of 35,000 2D velocity models with depth-wise gradients representative of Icelandic crust, perturbed by up to 25% with anti-persistent Von Kármán series and 1000 m correlation lengths. About 15% of these models contain fine-scale geological ‘dyke-like’ structures. Secondly, we forward model the wavefield gathers through each velocity model using SPECFEM2D, accounting for attenuation and broadband source properties. Thirdly, we train a Fourier Neural Operator (FNO) to predict 2D P-wave velocity models from single earthquake gathers. We show that FNO performance generalises to unseen earthquake gathers not included during training, recovering fine-scale velocity structure, including the dykes, from a single gather. In effect, this approach pushes the ‘big data’ requirement for the source side into the numerical domain used for training.

Notwithstanding challenges associated with field DAS instrument response variations, applying this approach to DAS data may open the possibility of high-resolution seismic imagery from a single to a few earthquakes in the field, and may have applications where DAS-enabled high-resolution body wave images can be obtained in regions with exceptionally low seismicity rates and where sufficient body waves cannot be extracted from ambient noise.