Novel applications of statistical techniques and machine learning in chemostratigraphy

Abstract:

Chemostratigraphy is a reservoir correlation technique involving the application of inorganic geochemical data. Data are typically generated for approximately 50 elements from each sample in the range Na-U in the periodic table using various established analytical instruments and techniques. Once acquired, data are plotted in the form of vertical profiles for each element and a range of elemental ratios. In most studies, profiles are plotted for a large number of parameters, more than 1000 combinations are possible, though chemostratigraphic schemes are typically based on variations in only 4-12 “key” elements or ratios. The role of the chemostratigrapher is to identify such key elements/ratios or other variables and parameters from each well or section before placing chemostratigraphic boundaries. Even with specific knowledge and skills this is an arduous and challenging process.

These challenges may be alleviated, at least in part, thorough automated methods that include statistical and machine learning techniques. The techniques of principle component analysis, quartiles, statistical boundary definitions and machine learning were found to be particularly useful in this regard. These methods were used in a number of studies presented here that could be applied on any subsurface reservoir unit. Statistical picking is helpful in identifying all major changes across all the variables and where important boundaries may be located. Principle component and quartile analysis are helpful in picking the key variables for stratigraphic analysis and it is demonstrated that variations in principle components, when plotted in profile form, can be utilized to identify correlative boundaries between wells. Discriminant function analysis can be employed to determine the depths at which boundaries should be placed in each well based on the statistical evaluation of the geochemical dataset. Case studies are presented to demonstrate how these statistical and automated techniques can be used in a multidisciplinary workflow to propose robust chemostratigraphic scheme. The impact of automations in chemostratigraphy is profound in terms of time-saving, and stratigraphic interpretation, as well as adding confidence to chemostratigraphic and geological correlations.