Radiometric method - Geophysics for the Mineral Exploration Geoscientist - page 206

Data acquisition Data reduction

Cosmic

Recorded spectra

Remove instrument
effects (partly)

Aircraft

Remove random noise by
spectral smoothing (NASVD, MNF)

Instrument
response

Atmospheric Rn

Remove aircraft &
cosmic background

K U Th

Remove atmospheric
Rn background

Correct for channel interaction

Scattering within
source & atmosphere

Correct for height-related scattering

Convert to ground concentrations of K, U & Th

Unknown quantities of K, U & Th in the ground Estimated quantities of K, U & Th in the ground

Figure 4.10 Factors contributing to the γ -ray spectra measured in the field and the corrections applied to reduce the survey data. The reduction
process is designed to resolve and count γ -rays emitted by radioelements in the ground beneath the sensor, and determine the local
concentrations of these elements.

summing where the photomultiplier combines the effects

of two nearly coincident radiation events and erroneously
records them as a single higher-energy event. These reduce
the instrument

- the data. The two most common methods are maximum

noise fraction (MNF) and noise-adjusted singular value
deconvolution (NASVD), both fairly recent developments.
The NASVD and MNF methods are both quite complex,
as are the arguments about their relative merits (Minty,
1998 ) . Both utilise principal components analysis (PCA),
which in turn requires pre-conditioning of the survey data
so that its statistical properties meet fundamental require-
ments of the PCA method (see Davis, 1986 ). The main
differences between the two methods relate to how this is
achieved, because both are based on the fact that the signal
is concentrated in the lower-order principal components,
whilst the uncorrelated (random) noise is concentrated in
the higher-order components. The smoothed spectra are
reconstructed using the lower-order components only,
thus removing the random noise. It appears that in many
cases similar results are obtained, but when they differ it is
a function of the survey data, as a high correlation between
U and Th affects the result. The important fact is that both
methods can signi

s ability to count scintillation events accur-
ately across the energy spectrum. There is no correction
available for these effects.

'

4.4.2 Random noise

A signi cant part of the noise component of a radiometric

measurement is random, caused by the random nature of
the radioactive decay process. This means there is no
correlation between the noise recorded in different energy
channels and, therefore, it can be suppressed using channel
correlation-based smoothing techniques, also referred to
as spectral smoothing. One approach is to treat the measured
spectra as though it were composed of several component
spectra. The challenge then is to identify these components
and reconstruct the measured spectra from each of them,
and in so doing identify the noise component. A more
widely utilised approach is based on statistical analysis of

cantly reduce errors in channel counts
compared with conventional processing: typically by 42%