Lundin Geolab’s AJAX ML model is Awesome!

Lundin Geolab’s AJAX ML model is Awesome! Lundin Geolab’s (AkerBP’s) trained Machine Learning models, which are available in OpendTect’s library of trained ML models, continue t...

Lundin Geolab’s AJAX ML model is Awesome!

Lundin Geolab’s (AkerBP’s) trained Machine Learning models, which are available in OpendTect’s library of trained ML models, continue to amaze us.

Today’s example is taken from a geothermal exploration project that we are currently working on. The data set is a legacy 3D dataset covering the city of Almelo in the Twente region of The Netherlands. Target is the pre-salt interval consisting of Permian and Carboniferous sands starting below the prominent deep reflector at around 1 s.

As you can see in the slider, the legacy data is, to put it mildly, rather noisy, and pretty difficult to interpret. DeSmile, the model that suppresses migration smiles, that we featured in this post two weeks ago https://lnkd.in/eZy7bjRN, does not work on this data set. The noise in this data set appears to be more random in nature. Instead, we applied AJAX, another 3D-Unet from the Lundin set of trained models.

AJAX not only tackles random noise; it also reshapes the amplitude and phase spectrum of the signal. The result on this data set is stunning. Seismic events post-AJAX are much more continuous and have more character, which greatly facilitates the interpretation in this challenging setting.

Author: Dr. Paul de Groot from dGB Earth Sciences #machinelearning #ml #seismicinterpretation #geoscience #geophysics #geology