On primaries-only travel times construction using Marchenko redatuming

Abstract

Extraction of primaries-only reflection data from full surface reflection data by a combina- tion of Marchenko redatuming and convolutional interferometry, incorporates artefacts, and noise following the constructed primaries. I have shown that the only true information recov- ered are primaries travel times, any dynamic information is artificial. I develop two extended versions of the original workflow, one of which processes seismic data throughout, avoids noise following the primary, and mitigates artefacts typical of this type of primaries construc- tion workflow. The other one uses travel time information as input and output, avoids noise following the primary, and also avoids artefacts by applying a quality control criterion dur- ing the convolutional interferometry step. It is cheaper and faster than the first version, and the resulting 2D primaries travel times matrices need less storage than seismograms. For complex data, it is difficult to extract a sufficient amount of kinematic information to gain an advantage over the full reflection data during velocity analysis and velocity model build- ing. The mechanics of the primaries construction workflow introduce a new class of artificial arrivals, which are indistinguishable from true arrivals. I have used imaging with the true model here to identify them. In addition, the bottom of a thin layer can be omitted in the final data, even if the primary reflection is present in the input data. I have used a horizon- tally layered subsurface model to show that mathematical convergence of the Marchenko redatuming scheme, as measured by the behaviour of convergence energy, is no guarantee for physical convergence towards the correct solution. Inverting for redatuming parameters using the trend of the convergence energy, or related measures applied to the same wave fields, can produce reliable looking results, but these are misleading and mostly false. Since the approach fails for a specific model, it is to be considered unreliable in general.