Rock classification from conventional well logs in hydrocarbon-bearing shale
Abstract
This thesis introduces a rock typing method for application in shale gas reservoirs using conventional well logs and core data. Shale gas reservoirs are known to be highly heterogeneous and often require new or modified petrophysical techniques for accurate reservoir evaluation. In the past, petrophysical description of shale gas reservoirs with well logs has been focused to quantifying rock composition and organic-matter concentration. These solutions often require many assumptions and ad-hoc correlations where the interpretation becomes a core matching exercise. Scale effects on measurements are typically neglected in core matching. Rock typing in shale gas provides an alternative description by segmenting the reservoir into petrophysically-similar groups with k-means cluster analysis which can then be used for ranking and detailed analysis of depth zones favorable for production. A synthetic example illustrates the rock typing method for an idealized sequence of beds penetrated by a vertical well. Results and analysis from the synthetic example show that rock types from inverted log properties correctly identify the most organic-rich model types better than rock types detected from well logs in thin beds. Also, estimated kerogen concentration is shown to be most reliable in an under-determined problem. Field cases in the Barnett and Haynesville shale gas plays show the importance of core data for supplementing well logs and identifying correlations for desirable reservoir properties (kerogen/TOC concentration, gas saturation, and porosity). Qualitative rock classes are formed and verified using inverted estimates of kerogen concentration as a rock-quality metric. Inverted log properties identify 40% more of a high-kerogen rock type over well-log based rock types in the Barnett formation. A case in the Haynesville formation suggests the possibility of identifying depositional environments as a result of rock attributes that produce distinct groupings from k-means cluster analysis with well logs. Core data and inversion results indicate homogeneity in the Haynesville formation case. However, the distributions of rock types show a 50% occurrence between two rock types over 90 ft vertical-extent of reservoir. Rock types suggest vertical distributions that exhibit similar rock attributes with characteristic properties (porosity, organic concentration and maturity, and gas saturation). This method does not directly quantify reservoir parameters and would not serve the purpose of quantifying gas-in-place. Rock typing in shale gas with conventional well logs forms qualitative rock classes which can be used to calculate net-to-gross, validate conventional interpretation methods, perform well-to-well correlations, and establish facies distributions for integrated reservoir modeling in hydrocarbon-bearing shale.