Page 214 - Geochemical Remote Sensing of The Sub-Surface

Light hydrocarbons for petroleum and gas prospecting 191

concentrations, regardless of location. A magnitude threshold, or series of thresholds, is
chosen and those locations with values above this threshold (anomalously-high
concentrations) are identified on a map. The second technique focuses on the spatial
clustering of anomalous stations. This is accomplished by the identification of regions
where the number of stations with magnitudes above a threshold is statistically
significant.
The traditional method of identifying a magnitude threshold has been accomplished
by a variety of techniques. These include: (1) the mean plus two standard deviations of a
normally-distributed data set; (2) arbitrarily selecting the 90 th percentile or 95 th
percentile, etc., of the data; (3) identifying the inflection point on a cumulative frequency
plot that deviates from a straight line (Sinclair, 1976).
In the opinion of the authors it is dangerous to select any hard-and-fast rule for
defining an anomalous population, although the approach of Sinclair (1976) is the most
appropriate for a mixed mode data set. Sample populations should be normal, or at least
log normal, for many of the statistical tests to be valid, and bias in sample sites should be
avoided if possible. Ideally a training set made up of a data subset with known
hydrocarbon potential should be employed. This gives a means to tie-in data to a known
feature, whether it be a source bed, a reservoir or a barren area. Once the results are
available, a first step is to construct histograms to determine the spread of the data. The
data can then be plotted on cumulative frequency plots to determine the different
populations. Scatter plots of key components, such as methane versus propane, or iso-
butane versus normal butane, often yield multiple trends for multiple populations in the
data. Pearson correlation analysis also yields useful information on the "cleanliness" of
the data, with single populations generally showing a high degree of inter-correlation.
Filtering or screening the data according to composition prior to applying statistics is
also an effective means of determining areas of favourable potential.
The method of identifying regions of anomalously-high leakage by clustering
(Dickinson and Matthews, 1993) is accomplished by first identifying a magnitude
threshold and a search area. The magnitude threshold, which is somewhat arbitrary, is
used to transform the distribution of magnitudes into a binomial population (above the
threshold "heads" and below the threshold "tails"). The size of the search area (the
"cell") is such that it includes 20 or more sample stations regardless of its location within
the surveyed area. Once these parameters are chosen, the cell is placed at one position on
the map, usually in one comer, and the percentage of heads and the total number of
stations within the cell are recorded. The cell is then translated to a new location and the
same parameters are recorded. This process is repeated until the entire survey area has
been examined. Because the properties of the binomial distribution are well known,
statistical tests of the chance of a particular cell having a particular percentage of heads
can be made and probability maps contoured. Thus, regions of anomalously-high
frequency of magnitudes above the threshold can be identified, and their chance of
arising due to random processes, instead of focused leakage, can be estimated. There is,