Impact of sedimentary facies on machine learning of acoustic impedance from seismic data

Speaker: Hongliu Zeng, Senior Research Scientist, Bureau of Economic Geology, The University of Texas at Austin This talk is focused on how to configure facies into machine learning (ML)-based seismic inversion. The main challenge is that most depositional systems have built-in complexities, which are difficult to describe in an ML model. Using a geologically realistic 3D model, I demonstrate that training score and prediction error can be correlated to facies pattern. ML with sparse wells is low score and highly unstable, which can be avoided by using a large synthetic training data set. Field-data tests show great potential to use ML in qualitative sand volume mapping, which can be helpful in studies of sedimentology, reservoir prediction, and CO2 sequestration.