GSC – NLGS Levelling Procedure
(from the Digital Geoscience Atlas of the Buchans –
Robert’s Arm Belt,
Although the same sample medium (the < 63 µm fraction of
till) was employed by Klassen and Liverman, different
analytical methods were used. Cross-analysis of a suite of 36 of Klassen's samples at the NL Geological Survey's geochemical
laboratory in
It should be emphasized that for mineral exploration, the relative variation of an element is of primary concern. Absolute accuracy is less important than precision, but if one set of results is systematically higher than another, false patterns due to these differences in accuracy will impair interpretation. To map geochemical variation over the entire belt it is necessary, therefore, to "level" data from the two sources for each element.
Compilation Method,
Combined Till Geochemical Database
Each author used two main methods of geochemical analysis
from a total of four laboratories; Inductively Coupled Plasma Emission
Spectrometry [ICP-ES] following acid digestion, and Instrumental Neutron
Activation Analysis [INAA]. In each study, the wet-chemical and INAA data are for a similar suites of elements. To evaluate the degree of
compatibility between data for common elements from each source, a set of 36
samples collected by Klassen were analysed
by the same methods and in the same geochemical laboratories that were employed
for Liverman's samples.
INAA is carried out on the dry sample and measures total element concentrations in most circumstances. The ICP-ES determinations for each study are based on quite different methods of digestion, however, and do not give directly comparable results for most elements. For Liverman's samples an HF-HClO4-HCl digestion was employed, which totally dissolves most samples and gives total concentrations of most elements. Klassen's samples were analysed by Chemex Labs Ltd., Vancouver, following digestion in hot, concentrated HCl-HNO3. This latter digestion does not completely dissolve the sample, and yields less than total values for most elements. Even where results are "total", discrepancies in calibration methods and other analytical factors can cause systematic biases between methods.
For each element, the results of each analytical method were compared through scatterplots and linear regression analysis. The scatterplots were used to identify erratic results from individual samples, which were then eliminated before computing the regression equations that are presented in Table 14. For this study, Klassen's results were levelled to match Liverman's because all Liverman's results are "total" abundance values that are calibrated by the inclusion of National Reference Materials. This does not imply that the methods used by Liverman are more precise than those of Klassen, however.
In compiling the combined till geochemical database, only elements whose regressions had a fit (R squared) of greater than 50% were considered. For several elements, e.g. Co, both determinations by Klassen's methods matched closely both determinations by Liverman's methods. For these elements, the match with the highest degree of fit was selected as the basis for computing the levelled variable in the combined till database. The regressions and element pairs selected for levelling are in italics in Table 14. In total, 28 elements showed a fit of > 50%, and many had much closer fits attesting to the good overall quality of both datasets. For those elements whose rgeression equation had a negative intercept, small negative values may result for low initial values. These have not been recoded; missing values are coded as -9. In the case of Au, because there was a significant scatter in the data, and the gold content of the reference material is not very well established, the data were directly merged without leveling.
The user should bear in mind the assumptions made when using the compiled data set. Because the levelling was done with a simple linear function, the relative variation within the data from each source is unchanged, even where the absolute values have been modified. [The original unlevelled Klassen dataset can be obtained from the Geofile Nfld/2611: Digital Geoscience Atlas of the Buchans-Robert’s Arm Belt]. See Table 15 for documentation of the descriptive variables.
Table 14.
Comparison of analytical methods used by Klassen
(1994)with those used by Liverman et al., (1995),
based on a set of 36 samples of <63 µm fraction of till samples analysed by all methods. Regression equations used to level
Klassen's data to Liverman's
are shown in italic.
Table 14a.
ICP methods: Regressions for results by Klassen's
method (HCl-HNO3 digestion) on results by Liverman's
method (HF-HClO4-HCl digestion), with 1 outlier removed.
Element Correlation % fit Significance Intercept
Slope
(R) (R2) (p)
Ni
0.97824 96 0.0000 2.90816 0.98986
Cu 0.97088 94 0.0000 0.44176 1.06889
Mn 0.95751 92
0.0000 360.46213 0.91750
Zn 0.94622 90
0.0000 22.49836 0.88077
Co 0.94597 89
0.0000 2.71311 1.01783
Pb 0.91745 84
0.0000 5.69301 0.75644
Cr 0.88220 78 0.0000 8.98213 1.34806
Fe 0.88216 78 0.0000 1.26023 0.84364
V 0.81799 67 0.0000 10.82591 1.56254
Ca 0.80417 65 0.0000 -0.10451 3.03483
Mg 0.79550 63 0.0000 0.34627 0.78734
Sc 0.78624 62 0.0000 7.95521 1.05593
********************************************************************
Regressions
statistically significant, but fit < 50%.
Al 0.69043 48
0.0000 5.41347 0.45338
Mo 0.68655 47
0.0000 0.46908 0.81635
Ti 0.65174 42
0.0000 3165.02736 1.52851
Sr 0.63360 40
0.0000 51.56729 4.24530
La 0.54105 29
0.0007 19.13579 0.38834
Na 0.50561 26
0.0017 1.62722 39.90788
K 0.43871 19 0.0074 1.72677 -5.19821
Ba 0.361310 14
0.0271 374.54079 0.65692
********************************************************************
Regressions not significant at
the 95% confidence level (p > 0.05).
Be 0.12469 2
0.4687
Ga 0.13468 2 -0.4336
---------------------------------
Table 14b.
INAA methods: Regressions for results by Klassen's
method (Chemex Labs) on results by Liverman's method (Becquerel Labs). "Prime"
symbols (' and ") indicate that one or two outliers, respectively, were
been removed prior to calculation of regression equation, based on a visual
inspection of scatterplots. Regression
are for Klassen's results on Liverman's.
Elem Correlation % fit Significance Intercept Slope
(R) (R2)
(p)
As 0.99489 99 0.0000
-0.80734
0.98892
Br 0.98232 96 0.0000
-0.18053
0.87854
Sb 0.97420 95 0.0000 0.01448
0.91072
Hf' 0.97244 95 0.0000 -1.28476
0.99116
Co 0.95547 91 0.0000 -0.66997
1.15394
Cr 0.92873 86 0.0000 8.22258
0.64814
Sc 0.92738 86 0.0000 0.71602
0.81795
Lu' 0.92621 86 0.0000 -0.11368
0.89156
Na 0.92407 85 0.0000 -0.18801
1.05503
Fe 0.90581 82 0.0000 0.40161
0.83094
Yb 0.89672 80 0.0000 -0.41349
0.75200
Th" 0.88841 79 0.0000 0.82432
0.79278
Cs 0.84759 72 0.0000 0.11230
0.82252
Rb' 0.82250 68 0.0000 10.92326
0.76660
U' 0.81319 66 0.0000 0.77646
0.56845
La 0.81088 66 0.0000 4.34869
0.94418
Sm 0.75075 56 0.0000 0.80418
0.49792
Au' 0.73039 53 0.0000 0.44077
0.59725
Ba 0.70831 50 0.0000 133.63799 0.64185
********************************************************************
Regressions
statistically significant, but fit < 50%.
Ce 0.66018 44 0.0000 18.50817
0.51608
Mo 0.51785 27 0.0080 0.48998 0.26043
W 0.47809 23 0.0032 0.39048
0.30476
Ta 0.46907 22 0.0039 0.67304
0.31546
Tb' 0.44001 19 0.0082 0.63816
0.29299
Ni 0.40110 16 0.0153 16.99855
0.12019
Eu 0.39528 16 0.0170 0.03941
0.66765
-------------------------------------------------
Table 14c.
Comparison of INAA method from Klassen with the ICP
method used by Liverman, with 3 outliers removed. Regression
are for Klassen's results on Liverman's.
Elem Correlation % fit Significance Intercept Slope
(R) (R2)
(p)
Co 0.98653 97 0.0000 -0.14891
1.18183
Na 0.96314 93 0.0000 -0.12104
1.10116
Sc 0.97738 96 0.0000 0.81060
1.01123
Fe 0.98056 96 0.0000 0.39017
0.95584
Cr 0.90600 82 0.0000 9.90677
0.52658
La 0.85518 73 0.0000 6.76356
0.77241
Ba 0.83125 69 0.0000 133.46092 0.64202
Ce 0.76858 59 0.0000 18.57977
0.73688
********************************************************************
Regressions
statistically significant, but fit < 50%.
Ca 0.67189 45 0.0000 0.49998
0.55016
********************************************************************
Regressions not significant at
the 95% confidence level (p > 0.05).
Ni 0.25689 7
0.1490
Mo 0.23736 6
0.2755
----------------------------------------------------------
Table 15. List of element variables, composite till geochemistry database in
the COMBINED dataset.
Variable unit Analyt. Analyt. Method of joining data
Method Method
Liverman Klassen
As_r_ppm ppm INAA INAA Levelled
by regression
Au_d_ppb ppb INAA INAA Direct merge, no levelling
Ba_r_ppm ppm ICP INAA Levelled
by regression
Br_r_ppm ppm INAA INAA Levelled
by regression
Ca_r_pct
% ICP ICP Levelled by regression
Ce_r_ppm ppm ICP INAA Levelled
by regression
Co_r_ppm ppm ICP INAA Levelled
by regression
Cr_r_ppm ppm INAA INAA Levelled
by regression
Cs_r_ppm ppm INAA INAA Levelled
by regression
Cu_r_ppm ppm ICP ICP Levelled
by regression
Fe_r_pct
% ICP INAA Levelled by regression
Hf_r_ppm ppm INAA INAA Levelled
by regression
La_r_ppm ppm ICP INAA Levelled
by regression
Lu_r_ppm ppm INAA INAA Levelled
by regression
Mg_r_pct
% ICP ICP Levelled by regression
Mn_r_ppm ppm ICP ICP Levelled
by regression
Na_r_pct
% ICP INAA Levelled by regression
Ni_r_ppm ppm ICP ICP Levelled
by regression
Pb_r_ppm ppm ICP ICP Levelled
by regression
Rb_r_ppm ppm INAA INAA Levelled
by regression
Sb_r_ppm ppm INAA INAA Levelled
by regression
Sc_r_ppm ppm ICP INAA Levelled
by regression
Sm_r_ppm ppm INAA INAA Levelled
by regression
Th_r_ppm ppm INAA INAA Levelled
by regression
U_r_ppm ppm INAA INAA Levelled
by regression
V_r_ppm ppm ICP ICP Levelled
by regression
Yb_r_ppm ppm INAA INAA Levelled
by regression
Zn_r_ppm ppm ICP ICP Levelled by regression