Data-driven Layer-stripping Strategy for 3-D Joint Refraction and Reflection Travel-time Tomography

We present and critically evaluate a data-driven strategy to successively invert different layers with joint refraction and reflection travel-time tomography. We use TOMO3D, a new 3-D travel-time tomography code that we have developed, which retrieves the velocity distribution and the geometry of reflectors based on refraction and reflection arrival times. Travel times are computed using a ray-tracing technique combining the graph and bending methods, and the LSQR algorithm performs the iterative inversion. The forward problem solution, which takes by far most of the run time, is parallelised. For practicality reasons, the code is designed to handle a single reflector per inversion, and it is combined with a layer-stripping strategy in cases involving several reflectors. Layers are defined by consecutive reflectors, and the final velocity model is built layer by layer, sequentially extending it with each new inversion and sorting data by offset. This strategy allows us to introduce strong velocity contrasts associated to reflectors, reduces the negative effect of velocity–depth ambiguity, and poses simpler inverse problems. Layer stripping is applied to a complex synthetic case including two curved reflectors. We show that this strategy allows to recover the velocity of both layers and the geometry of intermediate and bottom reflectors.