About This Item

Share This Item

The AAPG/Datapages Combined Publications Database

AAPG Bulletin


AAPG Bulletin, V. 87, No. 8 (August 2003),

P. 1377-1389.

Copyright copy2003. The American Association of Petroleum Geologists. All rights reserved.

Estimation of missing logs by regularized neural networks

M. M. Saggaf,1 Ed L. Nebrija2

1Saudi Aramco, P.O. Box 9863, Dhahran, 31311, Saudi Arabia; email: [email protected]
2Saudi Aramco, P.O. Box 12323, Dhahran, 31311, Saudi Arabia; email: [email protected]


Muhammad Saggaf has a B.S. degree from King Fahd University of Petroleum and Minerals and an M.S. degree and a Ph.D. from Massachusetts Institute of Technology. He joined Saudi Aramco in 1989 and has worked in exploration, reservoir characterization, and research and development. Along the way, he was awarded the Technology Achievement Award in 1997, the Creative Contribution Award in 2001, the SEG J. Clarence Karcher Award in 2001, and a patent for fractal deconvolution in 2002. He is a member of the Society of Exploration Geophysicists and AAPG, an elected member of the European Academy of Sciences, and the president of the Dhahran Geoscience Society.

Ed Nebrija has a B.S. degree in physics from the University of the Philippines and a Ph.D. in geophysics from the University of Wisconsin-Madison. From 1979 to 1992, he worked for Shell Oil (United States) in various capacities as marine seismic party chief, explorationist, and reservoir geophysicist. He is currently a consultant geophysicist at Saudi Aramco, specializing in the reservoir characterization of offshore and onshore oil and gas fields. He is a member of the Society of Exploration Geophysicists and the European Association of Geoscientists and Engineers.


We thank Saudi Aramco for supporting this work and for granting us permission for its publication.


An approach based on regularized back-propagation neural networks can be used to estimate the missing logs, or parts of those logs, in wells with incomplete log suites. This is done by first analyzing the interdependence of the various log types in a training well that has a complete suite of logs, and then applying the network to nearby wells whose log suites are incomplete to estimate the missing logs in these wells. The accuracy of the method is evaluated by blind tests conducted on real well-log data. These tests indicate that the method produces accurate estimates that are close to the measured log values, and the method can thus be an effective means of enhancing limited suites of wire-line logs. Moreover, this approach has several advantages over the ad hoc practice of manually patching the missing logs from the complete log suites of proximate wells because it is automatic, objective, completely data driven, inherently nonlinear, and does not suffer from the overfitting difficulties commonly associated with conventional back-propagation networks. Additionally, it seems that an accurate selection of the optimal input log types is not necessary because redundant input containing several logs yields reasonably accurate results as long as some of the logs in the input are sufficiently correlated with the missing log.

Pay-Per-View Purchase Options

The article is available through a document delivery service. Explain these Purchase Options.

Watermarked PDF Document: $14
Open PDF Document: $24

AAPG Member?

Please login with your Member username and password.

Members of AAPG receive access to the full AAPG Bulletin Archives as part of their membership. For more information, contact the AAPG Membership Department at [email protected].