Common conversion point stacking of receiver functions versus passive-source reverse time migration and wavefield regularization

We demonstrate with synthetic and field data that with sufficiently dense sampling waveequation-based methods such as reverse time migration (RTM), implicitly forming array receiver functions (ARFs), perform better resolution wise than migration of common conversion point (CCP) stacks of traditional receiver functions.However, even with modern array deployments the sampling requirement is typically not met for teleseismic (earthquake) data.To enable RTM imaging with sparsely (and irregularly) sampled wavefields at the surface, we use an intermediate reconstruction based on sparsity promoting optimization using a curvelet (or wave packet) representation of the data, as an important and necessary pre-processing step.To suppress artefacts, the curvelet coefficients are constrained to represent the range of known directions present in the data.We show that our proposed pre-processing procedure (which may be viewed as generating 'missing' traces) can produce artefact-free data for RTM even if only 20 per cent of necessary data are available in the original data set.With synthetic data, we also demonstrate that if the sampling criteria is not met, CCP can produce results that are superior over wave-equation methods such as RTM.As a proof-of-concept with field data, we image the structure of the crust beneath the Himalayas with passive-source RTM of teleseismic data from Hi-CLIMB project.For Hi-CLIMB data, the CCP and RTM results are similar because sampling is still too sparse for RTM and the structure is simple enough for successful CCP.Both results are improved by wavefield regularization and reveal that the Moho is continuous beneath most of the array, and not fragmented as suggested by some earlier studies.