Integrating lithofacies and well logging data into smooth generalized additive model for improved permeability estimation: Zubair formation, South Rumaila oil field

Integrating discrete distribution of lithofacies into the petrophysical property modeling is essential to preserve reservoir heterogeneity and improve flow modeling. Specifically, various studies have been implemented to model the permeability as a function of well logging data without taking into account the effect of lithofacies, which is rational to produce distinct regression lines given each facies type. In this paper, advanced statistical learning approaches were adopted as an integrated workflow to model the core permeability given well logging records and discrete lithofacies classification for a well in the South Rumaila oil field, located in Iraq. In particular, the probabilistic neural networks was first applied for modeling and predicting the discrete lithofacies classification given the well logging records: neutron porosity, shale volume, and water saturation. Next, smooth generalized additive model (SGAM) was used to model the core permeability as a function of the same well logging records. In addition, the predicted lithofacies was included as a discrete independent variable in the core permeability modeling to provide different regression lines given each lithotype. The SGAM was also modeled for three subset data given each separate lithofacies to verify the efficiency of SGAM and to provide more accurate prediction of permeability. The same procedure of SGAM was completely repeated by the generalized linear model (GLM) to prove the higher effectiveness of SGAM for permeability modeling and prediction. The root mean square prediction error in SGAM was lower than in GLM in all the combined and separate lithofacies models. In addition, the SGAM model overcame the multicollinearity between shale volume and water saturation variables by using the smoothed terms. Finally, making accurate permeability prediction for all wells in the field should ensure capturing the spatial variation and correlation between the data and then preserving the reservoir heterogeneity.