Specifications

University of Pretoria etd – Combrinck, M (2006)
noise or geological noise. Whatever the cause, noise in data will produce noise in the
processed product and in extreme cases may even cause a technique to fail. The S-layer
differential transform is especially sensitive to noise because of the numerical
differentiations performed. In addition to the optimal smoothing of data (chapter 4) a
filter is also applied to
remove noisy data points from each sounding before the transform
is applied. Noise in a sounding curve is easily recognised through visual inspection of
data, especially if presented in the logarithmic domain, by the sometimes random decay
behaviour of the later time channels. A filter mimicking this visual inspection is achieved
by computing the straight line regression coefficients for trios of data points starting from
the last three channels and working back to the front. When the regression coefficient is
larger than a specified value (0.997 default) it means that data conform to late time decay
behaviour. (This test is performed in both the logarithmic and semi-logarithmic
domains.) All data points after this channel are discarded as noise. Supplementary to
that, the first appropriate positive data value is chosen as the first point for each
sounding. This is done to negate the effect of sign changes sometimes manifesting in the
early channels. Examples of the application of this filter as well as the “S-layer differential
transform compatibility” (SDTC) filter (section 4.4.4) are shown in Figures 5-3 to 5-7.
103