Page 8 - i1052-5173-27-7

Table 2. Accuracy of melt redox compositions. In the field or in most microanalyses, ana-
lytical spots are chosen by the operator.
BAS-2 n = 1 n = 10 n = 50 n = 1 n = 10 n = 50 Therefore, the distribution of sampling
points must be strategically selected to
Beam/order n/a 0.1 0.1 0.1 950 950 950 minimize both the number needed and the
size ratio* sampling bias (Fig. 6). This can be
accomplished in one of two ways: analyze
SiO2 49.73 (0.35) 100.00 20.00 45.00 50.88 49.95 49.96 a large number of truly randomly selected
Al2O3 15.51 (0.19) 0.00 28.00 13.00 14.56 16.00 16.00 points, or estimate the mineral mode and
TiO2 (0.03) 0.00 apportion analysis locations to represent
FeO 1.55 (0.43) 0.00 0.00 0.00 1.92 2.03 2.03 each one appropriately (i.e., Chayes,
8.51 10.00 15.60 8.00 9.00 8.97 1956). The former is generally far easier
than the latter, with the modal analysis
Fe2O3 1.29 (0.03) 0.00 0.00 0.00 1.44 2.02 2.00 method presenting the additional potential
MgO issue of assuming a 2D surface mode
7.10 (0.08) 0.00 32.00 15.00 9.60 8.00 8.02 characterization represents a 3D rock
sample. New software for quantitative
CaO 11.48 (0.07) 0.00 10.00 6.00 10.40 10.01 10.00 EPMA mapping, which can provide full
compositional quantification of each
Na2O 2.66 (0.07) 0.00 0.00 5.40 3.20 2.99 3.02 image pixel (e.g., Carpenter et al., 2013;
Total 100.00 100.00 100.00 100.00 Carpenter, 2016), may reduce sampling
97.83 100.00 100.00 bias and better account for geochemical
heterogeneity.
Fe2O3/FeO 0.152 0.000 0.000 0.180 0.224 0.223
For rocks with distinct foliation, linea-
*Here order means the scale of the composition variation, which for a glass would be the scale tion, grain preferred orientation, or layer-
ing, sampling strategy becomes even
of long- or short-range ordering. more important; in these samples, an
accurate bulk rock composition may not
texture (phase heterogeneity) of the target Target Rock Type be meaningful. For example, if a rock is
rock, distribution of textural features layered, sampling traverses should cross-
within each target, sampling size of the It is intuitive to understand how a beam cut bedding planes with a sampling inter-
analytical instrument used, and sampling that samples the maximum number of val smaller than the layering interval to
strategy employed (Fig. 6). grains in proportions representative of the ensure proportional representation of
entire rock will yield optimal results. In each layer. Alternatively, traverses that
Figure 6. Comparison of the number of analyses coarser-grained rocks with varying abun- probe a single layer laterally may be
required for reproducible bulk compositional dances of minerals in their modes, sam- extremely useful, especially if multiple
data as a function of the scale of sample to pling strategy becomes critical; it is very layers are similarly studied for contrast.
measurement heterogeneity (upper). The same important that the phase assemblage be
plot is used to indicate where various analytical sampled proportionately. This may require Number of Analyses Needed
techniques and common geological samples plotting out sampling grids prior to analy-
might intersect (lower). STEM-EELS—scanning sis or point-counting phases on an outcrop Several sampling strategies can ensure
transmission electron microscopy–electron to determine the major phenocryst concen- quality analyses with reproducible results
energy loss spectroscopy; XAS—X-ray absorp- tration. In truly coarse-grained rocks (i.e., (Fig. 6). First, the larger the ratio between
tion spectroscopy; LIBS—laser-induced break- Fig. 4D), obtaining bulk compositions grain and beam size, the more analyses
down spectroscopy; EPMA—electron probe from smaller scale analyses is simply not are required. For grain sizes << beam
microanalysis. feasible. However, in such samples indi- size, 6–10 analyses produce a statistically
vidual mineral compositions may be repre- meaningful result as long as the phases
sentatively sampled, although fine-scale present are sampled proportionally
zonation might be obscured. (Figs. 4A–4C) and measurement accuracy
is appropriate. When the scale of grain
Sampling strategy is also critically size or the extent of ordering is close to
important when the analytical instrument or exceeds beam size (Fig. 4D), signifi-
has a sampling size much smaller than the cantly more analyses are needed (up to
crystallinity or long-range ordering of the 1000) to generate reproducible bulk com-
phase. As modern instrument resolution positional data. When more than one
continues to increase, understanding of phase is present, analysis locations must
sampling strategy will become even more be designed to represent all phases in the
critically important. rock proportionately if a true bulk rock
analysis is desired.
It is less obvious that relative chemis-
tries of the individual phases being studied
are important; if even one phase has dra-
matically different elemental abundances
over the other(s), then a larger number of
analyses will be needed to represent the
bulk. On the other hand, an ultramafic
rock composed solely of olivine and pyrox-
enes might have much less elemental vari-
ability among phases, and thus require
fewer analyses to be representative.

Distribution of Analysis Spots

Our model assigns sample locations ran-
domly to prevent systematic sampling bias.

8 GSA Today | July 2017