Towards realistic imaging and FWI testing

Daniel O. Trad, Ivan Sanchez

Research in academia often suffers from a limitation in the number of data sets employed for testing, resulting in a lack of feedback diversity that is crucial for comprehensive analysis. Applied research necessitates engagement with a broad spectrum of datasets, which significantly enriches research and development projects. Obtaining authentic datasets for publication in academia is not only challenging but also involves time-consuming preprocessing, making the pursuit of testing diversity a formidable task. Consequently, numerous tests are conducted on modelled data, often generated using similar algorithms employed in inversion processes, thereby giving rise to the "inverse crime scenario". The predominance of synthetic data testing in academia also comes as a consequence of the substantial difference in computational resources with industrial environments. Software developed in academia often lacks the capability to handle intensive computations with large seismic files with irregular acquisitions, capabilities that are required to work with real data sets used in industry. The consequence is a large gap between toy examples used in academia and realistic examples required for industrial use. This report details the implementation advancements made in our seismic libraries, showcasing tests aimed at enhancing the reliability of results in diverse environments, including large models, salt environments, topography settings, and physical models. Furthermore, we elucidate the disparities between inverse crime scenarios and realistic situations. The immediate ramifications of these advancements include the ability to circumvent the inverse crime problem and conduct tests in a variety of environments. Moreover, we anticipate that this research will foster increased collaboration with industry and deepen our understanding of the practical capabilities of novel techniques developed at CREWES.