DETERMINING THE LINE OF CORRELATION USING GENETIC ALGORITHMS

The most generally used method for estimating the global sequence and spacing of first and last occurrences, based on their occurrence in local sections, is the graphic correlation method of Shaw (1964). The key step in this method is the determination of the line of correlation (LOC), which represents the best estimate of the correlation between two local sections, or between a local section and a composite standard. In general, available techniques for fitting the LOC are tedious, subjective, or are computationally expensive. We will present a new method, genetic correlation (GC), that can dramatically reduce the effort involved in determining the LOC and produces stable biostratigraphic correlations and composite range charts objectively and efficiently. This approach is based the use of genetic algorithms, an artificial intelligence technique which excels in locating the optimum solution from large number of alternative choices. In the case of the LOC, the alternative choices are the number of line segments comprising the complete line and the positions of each segment's beginning and end points. Each option for the LOC is evaluated by a rule called the fitness function, which is based on a set of defined optimization criteria. GC uses an optimization criteria based on a least squares goodness-of-fit of the line segments to first and last occurrence datums (FAD's and LAD's). These fits are weighted by an estimate of the reliability of the FAD's and LAD's, using the assumption that the datums are more reliable for taxa with many occurrences within the range. A wide range of alternative LOC's can be rapidly evaluated and a potential optimum fit determined. By using the fitness function, it is also possible to estimate the point when no further refinement of the fit is necessary. The use of genetic algorithms to estimate the LOC is an order of magnitude more efficient than simulated annealing, an alternative artificial intelligence method.