Improving Pulsed Neutron Gamma Ray Spectroscopy Elemental Weight Percent Estimates through Automatic Dimensioning of the Spectral Fitting Process

Abstract

ABSTRACT Simultaneous measurements of neutron capture, activation, and natural gamma rays with the Geochemical Logging Tool (GLT*) string supply the necessary information to establish Si, Ca, Fe, S, Gd, Ti, Al, and К concentrations in formations. From these elemental concentrations, a reasonable geochemical description of many formations can be constructed. Such a description can be very useful to geologists and reservoir engineers in their efforts to understand the geochemistry of source and/or reservoir rocks in exploratory and development wells. Aluminum and potassium are measured in terms of weight percent; post-acquisition processing of the neutron capture data provide the Si, Ca, Fe, S, Gd, and Ti elemental concentrations. Normally, the post-acquisition phase is split into two parts: 1) a least-squares fit to extract uncalibrated elemental yields from the measured neutron capture gamma ray spectrum and, 2) a renormalization of the yields in a geochemical closure model to convert them into weight percent units. The latter of these steps eliminates the difficulty of determining the proportionality constant between an element's concentration and its fractional contribution to the measured spectrum. The discussion and methods presented in this paper focus on the spectral fitting procedure and its impact on the precision of the elemental concentrations obtained in the final step of post-acquisition processing. Large sections of log are usually fit for the subset representing the union of elements expected in the interval. However useful this practice might be for detection purposes, it can lead to erroneous and inaccurate elemental concentrations by including unnecessary degrees of freedom. More precise estimates of weight percent concentration can be realized by eliminating unimportant elements from the least-squares fit; the improvement depends on the number of elements eliminated and their correlation with those that are not. Founded on an analytic hierarchy, this novel approach produces results that compare favorably with those from more rigorous methods such as stepwise-elimination multiple regression. The proposed method is a 2-step procedure that does not require log analyst input. After performing a spectral fit for the union of elements anticipated in the interval, it automatically selects the proper subset of elements, and a refit of the measured spectrum is made at each depth. Adaptive dimensioning of the least-squares spectral fitting process in this manner reduces the uncertainties associated with the yields without sacrificing important information. This method can benefit formation evaluation in two ways. Because of its computational efficiency and ease of implementation, a real time application would increase the proficiency of geochemical logging. Second, a general improvement of elemental weight percent estimates over a broader range of lithologies is obtained, leading to better log-derived representations of subsurface formations.