Perpustakaan Digital ITB

Reverse time migration (RTM) is widely used due to its ability to recover complex geological structures that might not be imaged correctly with methods based on ray theory. However, RTM incurs a considerable computational cost. For this reason, we applied a model reduction technique that solves local spectral problems on a fine grid to simulate wave propagation rapidly on a coarser grid. We can also greatly reduce the computational cost by applying a frequency-domain implementation that requires only one matrix inversion for wave simulation. The multiscale model reduction, a fundamental aspect of the generalized multiscale finite element method, is obtained by utilizing a set of basis functions determined from a local spectral problem. We can tune the accuracy and computational speed by varying the number of basis functions. Applying RTM with fewer basis functions provides results more rapidly, though the usable results are restricted to lower frequencies. However, we can still clearly interpret large-scale structures from the multiscale RTM result. Tuning accuracy and speed of the imaging algorithm is especially useful when the velocity model needs to be actively updated, as it allows faster generation of trial images. In addition, multiscale RTM with a larger number of basis functions can provide faster imaging than the reference case, which uses the continuous-Galerkin finite element approach, without sacrificing accuracy.