Rapid Seismic Traveltime Modeling using Deep Neural Operator

Fast and efficient forward modeling is essential to the success of all inversion algorithms. Often, these inversion methods are bottlenecked by the computational cost associated with the forward solver. Despite decades of research on the topic, we still lack a robust and accurate forward modeling solver that can compute solutions instantly. Therefore, we introduce a neural operator-based method for rapid seismic traveltime modeling. We propose a novel framework to solve the factored Eikonal equation using the enriched deep operator network (En-DeepONet). Once trained, the network can be used to evaluate traveltime solutions corresponding to new source locations and velocity models instantly. Our results show that we can obtain highly accurate solutions instantly by using the trained network. This opens the door to quantifying uncertainty associated with seismic inverse problems.