TEDESCO, STEVEN A., and CHUCK GOUDGE, Atoka Exploration, Denver, CO

ABSTRACT: The Application of Pattern Recognition Methods in Surface Geochemistry

Statistical methods are currently the primary tool for evaluating surface geochemical data. Because the majority of surface geochemical surveys have been small, often less than 200 samples, this approach has been necessary. Although statistics on small data sets often correctly determines background and anomalous values they also can produce misleading information due to data set biases. Many of the statistical methods currently applied to petroleum surface geochemistry have come directly from the surface geochemical exploration for base and precious metal deposits. Although the two fields are similar in methodology and application they are measuring very different statistical populations. The geochemical "anomalies" in mining, like the ore deposits themselves, are normally a very sma l percentage of the overall population. Anomalies associated with petroleum deposits, however, can represent a large part of the population, just as the petroleum accumulations themselves can represent a large portion of an overall area. Further, these populations can grade from background to "anomaly" without statistically definable sub-populations developing. When a geochemical data set is small, in the number of samples or the areal extent, the survey may contain serious biases. If the geochemical population is dominated by samples which have been affected by hydrocarbon seepage simple statistics using mean and standard deviation will not identify legitimate anomalies as such. Even probability plots or more complex statistical analysis can often not identify or adjust for a biased dat set. A seriously biased survey or one in which the "anomalies" fail to reach an arbitrary statistical cutoff point may be wrongly rejected. While statistically "anomalous" samples may only indicate a non-economic deposit.

Industries often become dependent on a particular approach. This has begun to happen with statistics and geochemistry. Some have almost deified the use of statistics putting the statistics above the data. Some claiming that if it does not "exist" statistically then it does not exist at all. One must remember that statistics are only one of a number of different methods for interpreting data. Another approach, which is used extensively in the geologic sciences, is called pattern recognition. Pattern recognition is used to identify repetitive cycles in stratigraphy, porosity, permeability, etc. that are not statistically "anomalous" but are nevertheless important. Pattern recognition methods can be successfully applied to surface geochemistry constrained by two important elements: (1) a equate and appropriate models and (2) a clear understanding of the population parameters as determined from the models and background surveys. The models are most effective if collected across petroleum accumulations prior to or in the early stages of discovery. The geochemical anomalies seen in relation to the models are then used as the basis for interpreting the prospective area. Examples will be presented from actual surveys from the Las Animas Arch, Forest City basin, and Michigan basin to illustrate the limitations of purely statistical evaluations, and expand on the application of pattern recognition as a significant geochemical interpretation method.

AAPG Search and Discovery Article #90987©1993 AAPG Annual Convention, New Orleans, Louisiana, April 25-28, 1993.