Low-frequency data extrapolation using a feed-forward ANN

Research output: Chapter in Book/Report/Conference proceedingConference contribution

28 Scopus citations


Full-waveform inversion (FWI) benefits in many ways from having low-frequency data. However, those are rarely available due to acquisition limitations. Here, we explore the feasibility of frequency-bandwidth extrapolation using an Artificial Neural Network (ANN) approach. The ANN is trained to be a non-linear operator that maps high-frequency data for a single source and multiple receivers to low-frequency data. Assuming that the source is a point (delta function) in both time and space, we train the network on synthetic data generated using random velocity models. Extending our previous work, we apply the ANN to multiple collocated source-receiver acquisitions to predict 0.5~Hz data for a crop from the acoustic BP 2004 benchmark model. Prediction results, in general, resemble the reference ones but the prediction accuracy is barely sufficient to directly use extrapolated data in FWI. To demonstrate, we show regularized mono-frequency FWI on extrapolated data.
Original languageEnglish (US)
Title of host publication80th EAGE Conference and Exhibition 2018
PublisherEAGE Publications BV
ISBN (Print)9789462822542
StatePublished - Oct 16 2018

Bibliographical note

KAUST Repository Item: Exported on 2020-10-01


Dive into the research topics of 'Low-frequency data extrapolation using a feed-forward ANN'. Together they form a unique fingerprint.

Cite this