Sonic log predictions using seismic attributes

Todor I. Todorov, Daniel P. Hampson, Brian H. Russell

Deriving a deterministic relationship between the seismic data and geologicalproperties of the subsurface is a difficult task. Using multi-regression analysis and neural networks, we derive statistical rather than theoretical relationships. The relationship is found at the well locations and applied to the exploration area covered by seismic data.

Nine well locations in the Blackfoot area, Alberta, are used to derive relationships between the measured sonic velocity and seismic attributes. Cross-validation tests are used to determine the quality of the derived relationships. Using a neural network we achieved the highest correlation between the measured and the predicted sonic logs: 0.87. 3-D sonic velocity volumes are generated and a low-velocity anomaly is interpreted as a sand channel.