Multivariate Analyses using Geologic Information from Cores (MAGIC).

CONTENTS

MAGIC PROJECT OVERVIEW  TOP

Introduction

Multivariate statistical analysis of geophysical log data generally use one or more of the following statistical techniques:

The literature provides several examples of discriminant analysis (Crowe & Dodman, 1989), principal components (Mulhern, et. al., 1988) and cluster analysis (Anxionnaz, et. al., 1990). All of these procedures for lithofacies identification will give roughly similar results given similar input data.

The Multivariate Cluster analysis program known as MAGIC has been applied to 364 wells in Southwest Alabama as part of a regional study of the Smackover Formation. MAGIC uses the data from wireline logs (360,000 feet of neutron, density, sonic, gamma ray and deep resistivity data) along with core data (47,000 feet of lithologically described and analyzed cores) to predict lithologies. The availability of modern vintage (i.e., post 1970) cored and logged wells in the study area is exceptional from a data standpoint. The density of this core and log information has provided experimental verification of the multivariate technique.

MAGIC is written using Statistical Analysis System (SAS) language. The statistical results of MAGIC log analysis are saved to external SAS datasets for further analysis and to provide input to other software packages for stratigraphic study and mapping.

Detailed predicted lithologies derived from MAGIC in the SAS-maintained datasets are used to identify favorable stratigraphic reservoir trends in complex carbonate and evaporite sequences.

Benefits of MAGIC Log Processing.  TOP

The following benefits are realized from using the MAGIC processing technique:

Uses of MAGIC Logs  TOP

Color MAGIC logs have been used to develop time-correlative sequence events in an area with no distinctive fossils or marker bed boundaries. The MAGIC logs are further integrated with core thin sections, field and production data, and map analysis

Multivariate log and core statistical analysis and seismic data provided  a comprehensive dictionary of Smackover data in Mississippi and Alabama. Since much of these data are digital, many tentative (what if ?) correlations can be made in short order.

Background Review  TOP

The traditional method of using chart book cross-plots is not suitable for determining correct rock petrology in areas of complex lithology. Statistical treatment of well log information improves lithologic predictions.

Procedures based on discriminant analysis use a truth table of log responses and core lithology. Unfortunately, dolomites and compacting shales have such a wide range of log values which makes the precise lithofacies identification of these rock types hard to perform.

Schlumberger Faciolog and Geocolumn (Mulhern, et. al., 1986) procedures are based on principal component analysis. Principal components are manually grouped into local facies nodes which determine lithofacies identifications. These are incorporated into their CLIFF (Core-Like Interpretation of Facies Features, Anxionnaz, 1990) experimental system. The results from these analyses cannot easily do multiwell population studies or provide quantitative exploration mapping of lithofacies distributions.

MAGIC is based on SAS code. As such it is an integrated system which runs on any system (PC, VAX, Main Frame, UNIX, and APPLE computers) and talks to many types of databases (Oracle, M204, VSAM, DB2, dBase, Sybase, Access, and others). The built in graphics, statistics, utilities, and database input/output allow rapid end user access to data analysis.

SMACKOVER REGIONAL PROJECT  TOP

The Smackover Regional project began January, 1988 to investigate the use of well log derived lithology data integrated with core information. Very little previous work existed in a form that allowed us to add additional lithologic data in a quantified manner. An accurate depiction of complex carbonate lithofacies is required for detection of subtle hydrocarbon traps in the Smackover Formation.

Log and Core Data Resources In Study Area  TOP

A cursory examination of all well data resources in Alabama shows that:

The large percentage of core and modern log information has provided the statistical base needed to prove the validity of multivariate lithology identification technique. Chevron has about 230 wells with extensive log suites of neutron, density, sonic data in the area. These data are available in technical well files.

State of Alabama Core and Thin Sections  TOP

A very important part of our study is based on the existence of thousands of thin sections archived at the Alabama State Geological Survey. The State of Alabama requires all core and logs taken from wells drilled within state borders be deposited at the Survey after a period of time. These core and core-chip specimens are available for public use. The only operational requirement is that a copy of the study results are given to the State. Many thin-sections made from core materials have accumulated over the years from all different sources, both industry and academic, and may be examined at the Alabama Geological Survey.

We have micro-photographed about 800 thin-sections from the Alabama collection. These photomicrographs are used to determine lithological facies and make interpretations on original environments of deposition. MAGIC logs helped us to determine what the rocks are now, the thin-sections help us to determine what the rocks were during deposition.

Digitized Data Used In Project  TOP

After analyzing the results from 364 wells, 190 of them with cores, 12 of them by personal description of the cores, and 35 of them with extensive thin-sections; Chevron now has a Smackover database containing 360,000 feet of normalized, depth adjusted induction, gamma ray, spontaneous potential, neutron porosity, density, sonic and caliper curves with 47,000 feet of core description plus core porosity and permeability data.

Results of First Generation MAGIC Analysis  TOP

This use of statistical log and core analyses has resulted in a significant improvement in understanding Smackover depositional processes as compared to prior in-house studies and other reports. This improvement in our understanding of Smackover depositional environments has allowed selection of many additional areas for hydrocarbon prospecting in Mississippi and SW Alabama.

Comparison of Lithologies from Core and MAGIC Analyses  TOP

Figure 1. Old MAGIC Results Using All Core Descriptions.

CORE DESCRIPTIONS MOSTLY FROM COMMERCIAL LABS

OVERALL AGREEMENT LOG TO CORE LITHOLOGY = 73.5%

Comparison of Lithologies from Core and MAGIC Analyses - Core Descriptins from Experienced Geologists  TOP

Figure 2. Old MAGIC Results Using Experienced Geologist Core Descriptions.

OVERALL AGREEMENT LOG TO CORE LITHOLOGY = 76.9%

DESCRIPTION OF THE MAGIC LOG PROGRAM  TOP

Brief Synopsis of Operating Logic

MAGIC log data processing consists of reading log and core data files from datasets containing the project files.  SAS reads in project file log data and carries out the following steps:

Bad Log Data Processing For Dead Trace  TOP

Usually, much of the 'bad' log data are identified using cut-off lines on neutron-density, neutron-sonic, and sonic-density cross-plot tables. Cut-offs for 'bad' log curve data recognized during this study are as follows:

ACCEPTED DATA RANGES FOR CURVES USED IN CLUSTERING

CROSS-PLOT DATA CUT-OFFS

The caliper curve was not used to determine 'bad' data. Many MAGIC logs have demonstrated that hole size along with hole rugosity has a great effect on density pad contact with the borehole surface. This generally shows up as low values of bulk density readings along with curve correction values in the high minus range. In this particular case density curve values may be invalid (i.e. 'bad') even though the sonic, neutron and induction curves show valid rock values. By using heuristic logic developed during the Smackover Study, offending curve(s) are set to null values for statistical curve reconstruction.

Statistical Reconstruction of Bad or Missing Data (DEAD TRACE)  TOP

SAS code member DEADTR executes a statistical reconstruction of all 'bad' or missing data by separating missing data into three temporary SAS data files; no-density, no-sonic, and no-neutron. For example, the no-sonic file has valid values for density, neutron, and deep induction curves with a missing sonic value. The no-density and the no-neutron files have valid values for other curves.

Statistical reconstruction of missing data starts by going to the SAS database containing log digits of all logs for all wells previously processed by MAGIC. From this multiwell log file, only 'good' (i.e., no bad or missing log curves) log data is selected. To further window the data to manageable size, only those 'good' data from identical formations in the missing data files are selected for statistical analysis.

For each SAS data file with missing curves, (no-density, no-neutron, and no-sonic cases) data are retrieved from the 'good' log data set by selecting only those values occurring in the same modal class as the 'missing' log data by using an identification (ID) variable:

The data in the 'good' log data are further reduced by calculation of the means, sums, and frequencies of all data in each good-log-data-set by using a ID variable. This reduces the population of 'good' log data to 1,000-to-10,000 points. This makes the DEADTR statistical routines which determine the estimated value of missing log curves cost effective. Since the ID variable is identical in both the 'bad' and the 'good' data, all good-log-data is merged with all "bad" log-data. The missing curve value is supplied by the 'good' dataset. Insufficient data in any one of the 'good' log sets nullify the DEADTR algorithm.

Two statistical procedures are executed for each case, no-sonic, no-density, and no-neutron:

The value for DEADTR no-curve estimate is calculated by averaging the GLM value with the FASTCLUS modal value to smooth out 'lumpy' modal values and make a more natural appearing curve.

MULTIVARIATE CLUSTER PROCEDURE  TOP

Restored Log Data After Dead Trace

Prior to using multivariate statistics, replacement DEADTR data for bad or missing curve values are merged with original log curve data. The use of member BADLOGS finds all 'bad' data points and sets those values to null values. This avoids clustering of data outliers and limits 'good' log data to the coverage of seed data values in 5 dimensional space. The set up calculation for all the clustering variables is then performed by member MSPEQ.

Member MSPEQ contains log based equations to calculate sonic-density, density-neutron, neutron-sonic xplot porosity and sonic and density matrix values. There are optional NPHI environmental calculations if needed.

All default values for log constants are as follows:

Neutron - Density Cross Plot Porosity  TOP

IF THEN DO;

PORDA=(RHOM-4.0)/(RHOM-RHOF);

PORNA=0.76-10**(-(5*NPHI+0.16));

END;

IF DO;

PORDA=1.0; PORNA=-(2.06*NPHI+1.17)+10**(-(16*NPHI+0.4));

END;

NDXPHI=(PORDA*NPHI-DPHI*PORNA)/(PORDA-PORNA);

Calculate Apparent Matrix Density  TOP

IF ABS(NDXPHI)¬>=1.0 THEN RHOMA=(DRHO-ABS(NDXPHI)*RHOF)/(1-ABS(NDXPHI));

ELSE RHOMA=1.0;

NPHI=NPHI*100;

DPHI=DPHI*100;

NDXPHI=NDXPHI*100;

Neutron - Sonic Cross Plot Porosity  TOP

IF NPHI>SPHI THEN DO;

PORSA=-.146;

PORNA=0.5-10**(-(5*NPHI+0.3));

END;

IF NPHI<=SPHI THEN DO;

PORSA=0.5; PORNA=-(.62*NPHI+0.36)+10**(-(18*NPHI+0.92));

END;

NSXPHI=(PORSA*NPHI-SPHI*PORNA)/(PORSA-PORNA);

Calculate Apparent Matrix Transit Time  TOP

IF ABS(NSXPHI)¬>=1.0 THEN DTMA=(DT-ABS(NSXPHI)*DELTF)/(1-ABS(NSXPHI));

NPHI=NPHI*100;

SPHI=SPHI*100;

NSXPHI=NSXPHI*100;

Sonic - Density Cross Plot Porosity  TOP

X=((RHOF-RHOM)*(DELTSH-DELTM))-((DELTF-DELTM)*(RHOSH-RHOM));

SDXPHI=(((DRHO-RHOM)*(DELTSH-DELTM))-((DT-DELTM)* (RHOSH-RHOM)))/X;

SDXPHI=SDXPHI*100;

This allows a prediction of lithology using the best possible input data to the FASTCLUS procedure.

FASTCLUS Clustering of Log Data  TOP

Calculation of Cluster Variables  TOP

Data preparation for clustering log data includes a calculation of the actual variables used in the clustering algorithms. The variables derived from log data are as follows:

The curve weighting factors used to generate clustering variables for FASTCLUS are calculated as follows:

* Weighting factor is calculated by dividing relative lithologic contribution by population values of the FASTCLUS variable at the 98% cumulative distribution point minus the value at the 2% cumulative distribution point. RHOMA and DTMA are weighted the most.

The remaining 'good' data are clustered using a disjoint clustering SAS procedure called FASTCLUS. FASTCLUS is used to find 2628 preliminary clusters. These clusters and the output data set are sorted in preparation for later merges with the cluster lithology key table. Seed values are generic for each cluster and have been chosen by using cumulative distribution frequencies of all log curve data.

Cluster Lithology Key Table  TOP

All 'good' log data are merged with the cluster-lithology key table by cluster. The cluster-lithology key table (With all non-parametric lithologic associations) contains cluster numbers and associated lithology assignments which have been determined by comparison to previous core intervals. The following items pertain to construction and use of the key table:

Lithology and Petrography from Log Data  TOP

Lithology is assigned by using key petrologic terms which are recognizable by later steps of the MAGIC analysis. These key terms are combined into a single variable called LITH. LITH is composed of a major and (if needed) a minor lithology. For example, a limey dolomite is identified to MAGIC as DOL,LS. Using major-minor pairs of lithologies, it is possible to pair up 92 separate lithologic types using the following basic rock types:

* Note - These lithology types are based on lithology symbols available to the plotting package which is used. Additional symbols could be used if more lithology symbols are available.

** If a basement formation call is made in the proprietary marker files or a BSMT call is made in the core files then all lithologies are changed to BSMT.

The major-minor lithologies from logs match major-minor nomenclature in core files.

These major-minor pairs are also identified in member SYMTABLE which contain the symbol and color instructions for preparing log displays with log and core lithofacies.

OUTPUT FILES  TOP

SAS writes data needed to create a color log copy. The procedure is described as follows:

MAGIC Database Maintenance  TOP

SAS and MSP code books in the References Cited section are needed to get the most out of this documentation. The first assumption made is that the reader is somewhat familiar with SAS.

SAS documentation is especially needed since file maintenance is done using interactive SAS. When using combined SAS file LOG data to do multiple well processing, much information can be gained by simple SAS procedures. For example:

Data File Allocations  TOP

MAGIC uses a process of statistically analyzing information contained in data files of logs and cores. There are many external files (SAS MAGIC source code, Project Data, Utility, and MSP JCL) used by MAGIC which need user maintenance. These are:

Project Files.  TOP

Project log data are copied from MSP TRACES or WELLLOG data into a flat file (Eg. EXL.R$MAJIC.AASSCCC.LOGS where AA=state, SS=state code, and CCC=county code) for MAGIC to read.

Project core data are entered from core reports and verbal core descriptions into a flat file (Eg. EXL.R$CORE.MAJIC) for MAGIC to read.

Utility Files.

Utility SAS datafiles (Eg. EXL.R$SAS.UTILITY) used by MAGIC for marker name translations and state-county names from API numbers. The general utility data does not change and is not maintained by MAGIC.

Marker date maintained by the user are important since they are used in reconstructing DEADTR intervals and annotation on track 1 of the MAGIC logs. Formation calls are used in MAGIC to annotate the MAGIC log and tag LOG digits for use in multiwell processing and DEADTR reconstructions.

The data is entered using PIDMS or GWDS proprietary NA card records. The study name are the only marker calls recognized by MAGIC. This name has the value of the SAS macro variable &STUDY. This value of the 5 character formation call nmonic is maintained downward from the point of first occurrence until another formation call is encountered. Formation data is downward propagating by depth in the log data.

SAS Active Data Files  TOP

SAS active data files are SAS datasets maintained by MAGIC. These datasets form a basis for multiwell processing of data. Since they are in SAS format it is necessary to learn certain SAS PROCs such as PROC FSEDIT to do any file maintenance.

SAS data maintained by MAGIC:

STARTING A NEW PROJECT  TOP

So far, in discussions of MAGIC programs, areas used in the examples are limited to the Haynesville, Smackover, and Norphlet in Mississippi and Alabama. While most of the multivariate observations loaded into member CLUSMKV are in these formations, there is evidence that a significant portion of these observations have world-wide significance. For example, anhydrite precipitated from a Permian sea is identical to anhydrite precipitated from a Jurassic or Cretaceous sea (given similar depths of burial).

Petrologic differences in lithologic rock types of a similar composition may be due to differences in local usage. Truly different rock types should be recognizable as unidentified clusters on a MAGIC log. Currently only half of the 2832 clusters in the FASTCLUS seed table are correlated to sandstone, evaporite, limestone, and dolomite lithologies found in the Jurassic of SW Alabama and Mississippi. The other half of all missing or unidentified clusters probably represent:

Because of the large number of clusters used in FASTCLUS, it may be impractical to start out with a clean cluster key table. Perhaps a better way would be to start out with a copy of CLUSMKV and change cluster assignments in the copy of the cluster-lithology key table as new log and core data show differences. Many of those cluster-lithology identifications now used for the Jurassic of SW Alabama in CLUSMKV may be valid for many areas in the United States.

Lithology Identification in New Areas  TOP

To examine the feasibility of using MAGIC in new areas, the following need to be determined.

LOADING LOG AND CORE DATA  TOP

Loading of log and core data forms the project database for the MAGIC program. You should be consistent in the internal layout of the project file. It is suggested that naming conventions on the project log files be followed as used in the Smackover Regional project.

Digitization of Log Data (Smackover Regional Project Example)  TOP

All data used in this study were digitized from hardcopy paper records because extensive digital files do not exist in SW Alabama. As a result, not all environmental corrections were applied to log data. Only simple log transformations were performed such as deep resistivity from conductivity, bulk density from density porosity, & etc.

These logs were initially examined by us and each log curve was color coded for digitizing. Color coded logs from each well were digitized and loaded into a WELLLOG database without any corrections, i.e. verbatim copies of each log:

Adjust and Normalize Log Data  TOP

Digital log data generally come from two sources:

After log curves are digitized then a replot of the curves are compared with the hardcopy records for quality control. The digital curve data is then loaded as recorded into the WELLLOG database for safekeeping and preservation of raw data.

The project file is loaded from the WELLLOG database by merging the following curves by depth:

These project files become the basis for creating MAGIC logs. Curves are normalized and depth adjusted as needed by inspection of the curve log plots. We have found that the following curve adjustments are often necessary. They are (in order of decreasing occurrence) as follows:

Since the MAGIC program only recognizes log data having a valid reference numbers and depths, the user may write notes in the top of each file in the project file to record changes made to data in each well for future reference.

Digitization of Core Data (Smackover Regional Project Example)  TOP

Core data was handled by keypunch from core reports found in wellfiles or obtained by scouting sources. Core descriptions were added to keypunch files if good proprietary data were found, otherwise contractor core descriptions were used. Keypunch files were concatenated together and contain the following data:

Depth Adjust Core Data  TOP

For the Smackover Regional project, core data was entered by local key punch or individual geologist. Core descriptions were obtained from in-house proprietary documents, personal inspections of core at the Alabama Geological Survey, published geologic literature, or lastly descriptions accompanying the commercial core report. Porosity and permeability values come from the commercial report.

We have found that the following core file adjustments are often necessary. They are (in order of decreasing occurrence) as follows:

REFERENCES CITED  TOP

Anxionnoz, H., Delfiner, P., and Delhomme, J.P., 1990: Computer-Generated Corelike Descriptions from Open-Hole Logs, AAPG Bull, V. 74, No. 4,p. 375-393

Mulhern, M. E., Laing, J. E., Widdicombe, R. E., Isselhardt, C., and Bowersox, J. R., 1986: Electrofacies Identification of Lithology and Stratigraphic Trap, in GEOBYTE, V. 1, No. 4, p. 48-56

SAS, 1979: Statistical Analysis System (SAS) Users Guide, Version 79.2, SAS, Institute Inc., Raliegh, NC.

SAS, 1982: Statistical Analysis System (SAS) Users Guide-Basics, Version, 82.4, SAS Institute Inc., Raliegh, NC.

SAS, 1985a: Statistical Analysis System (SAS) Users Guide-Basics, Version, 85.0, SAS Institute Inc., Raliegh, NC.

SAS, 1985b: Statistical Analysis System (SAS) Users Guide-Statistics, Version, 85.0, SAS Institute Inc., Raliegh, NC.

SAS, 1985c: Statistical Analysis System (SAS/GRAPH) Users Guide-Graphics, .Version 85.0, SAS Institute Inc., Raliegh, NC.

SAS, 1985d: Statistical Analysis System (SAS/FSP) Users Guide, Version 85.0, SAS Institute Inc., Raliegh, NC.

© 1999, 2000

Dave Barthelmy (daba@wt.net)

3607 Aldergrove Dr.

Spring, TX 77388-5002

 

(281) 350-4120  (home)

(281) 989 6052  (mobile)

(281) 288-9410  (fax)