Content uploaded by Dennis Darby
Author content
All content in this area was uploaded by Dennis Darby on Mar 31, 2016
Content may be subject to copyright.
Journal of Sedimentary Research, 2015, v. 85, 247–253
Research Methods
DOI: http://dx.doi.org/10.2110/jsr.2015.17
CHEMICAL FINGERPRINTING, A PRECISE AND EFFICIENT METHOD TO DETERMINE
SEDIMENT SOURCES
DENNIS A. DARBY, WESLEY MYERS, STEPHEN HERMAN,
AND
BRIT NICHOLSON
Department of Ocean, Earth, and Atmospheric Sciences, Old Dominion University, Norfolk, Virginia 23529, U.S.A.
e-mail: ddarby@odu.edu
A
BSTRACT
: An accurate method of determining source locations for detrital sediment is presented using the chemical composition of Fe-oxide
minerals like a fingerprint. This method is an improvement in the use of Fe-oxide minerals for provenance determinations because it requires less time
and fewer source samples. A rigorous test of the method uses a database of more than 38,000 grains from known locations. The average error of
matching grains back to 45 source locations designated for this database is less than 2%. The method allows for proportional matching of a grain to
multiple sources if other grains in the source database meet the compositional match criteria, which helps reduce the error of incorrect matches. Most
provenance studies do not involve source basins as large as the entire Arctic Ocean, where sediment can be ice-rafted several thousand kilometers. For
most studies, only a few samples (
,
100 grain analyses/sample) would be required to characterize a source if strategically placed, such as near a river
mouth. Deposits more than 40 million years old can be traced to specific sources using this method because Fe-oxide grains are relatively stable in
most deposits.
INTRODUCTION
Determination of provenance is one of the most important goals in
sedimentology because such information can answer fundamental questions
concerning the nature and direction of sediment transport, mixing of sources,
and deposition. Heavy minerals that normally make up less than two percent
of the minerals in a deposit have been used with limited success to determine
sources (e.g., van Andel 1959; Mange and Wright 2007). There has been more
success in revealing sources using the isotopic composition of rare earth
elements in some minerals, age determined from detrital zircons or other
minerals (e.g., Rasbury et al. 2012), or the chemical composition of mineral
groups like garnets (e.g., Morton et al. 2004). Typically, suitable quantities of
each of these detrital components are not available for rigorous statistic
testing. Despite being the largest group of accessory minerals in most
deposits, opaque minerals are commonly ignored, resulting in missed
information about provenance, source geology, and location (Blatt et al.
1972). This paper presents an improved method that uses the chemical
composition of many of these opaque minerals (i.e., the anhydrous Fe-oxide
minerals) (referred to as Fe-grains) for determining provenance. This new
method requires less time to determine precise sources than earlier uses of
these minerals (Darby 2003) and with less error of misidentified source
determinations.
Electron-probe microanalysis (EPMA) provides the chemistry of individual
grains and allows for determination of not only more precise sources but also
multiple sources that might contribute to a deposit when each grain is
matched to all potential sources (Basu and Molinaroli 1989; Darby 2003;
Darby and Bischof 1996a, 1996b; Bischof and Darby 1999; Darby and
Zimmerman 2008; Darby et al. 2001, 2002, 2011, 2012).
Fe-oxide grains have diverse chemical compositions that can be used in
source determination. For example, magnetite can accommodate up to 28%
substitution for Fe (37% FeO) by other elements, especially Zn and Ni (Lewis
1970; Waychunas 1991; Darby 1998; Suavet et al. 2009). When the EPMAs of
8,555 titanomagnetite grains from the circum-Arctic source database are each
re-summed to 100 to account for small differences in EPMA results primarily
due to minor differences in carbon coating between standards and samples,
the average sum of all analyzed elements except Fe and O is 12.7%
(Supplementary Materials, Table A1). We include Ti, which averages 9.4%,
because it is highly variable in this mineral (s54.5%). Other Fe-oxide
minerals also have significant substitutions in the mineral lattice of many
elements (Haggerty 1976; Gaspar and Wylle 1983; Cornell and Schwertmann
2003). Unaltered ilmenite is the most abundant Fe-oxide mineral in the
database at 42% of all analyzed grains; the average of all elements other than
Ti, Fe, and O in this mineral is 2%, but it can be as high as 21.9%
(Supplementary Materials, Table A1). EPMA analyses of Fe-oxide minerals
such as ilmenite, magnetite, titanomagnetite, hematite, and chromite as well
as ferro-ilmenite (ilmenite with ,50% exsolved hematite, Fig. 1), titano-
hematite (hematite with ,50% exsolved ilmenite), and magnetite with other
exsolved phases can contain measurable amounts of several elements besides
Fe, O, and Ti. These include Mn, Mg, Si, Al, Ca, Zn, V, Ni, Cr, Nb, and Ta.
Fe-oxide minerals are common in sand deposits. This is partly due to their
abundance in igneous and metamorphic rocks as well as their relative stability
and durability (Grigsby 1992; Craig and Vaughan 1994; Cornell and
Schwertmann 2003). Extensive leaching is required to alter ilmenite to
pseudorutile (Dimanche and Bartholome 1976), and most temperate climates
produce only mildly altered ilmenite (Force 1991). Hematite is more readily
altered, as evidenced by the depressions from previous exsolution lamellae in
ilmenite that is otherwise still fresh (Darby 1984; Fig. 1) and the near absence
of this mineral in beach deposits of Virginia and North Carolina while it is
abundant in local rivers (Darby 1984, 1990; Darby and Tsang 1987; Darby
and Evans 1992). Fe-oxide minerals are abundant in beach and shelf deposits,
even Pleistocene deposits throughout the east coast of North America (Darby
1990; Darby and Evans 1992). They are rarely altered in marine sediments, as
evidenced by their unaltered abundance in sediments as old as 44 million
years in the ACEX core (IODP site 302) from the Lomonosov Ridge, Arctic
Ocean (Darby 2008, 2014). Of the more than several thousand samples that
we have processed, fewer than 1% did not contain adequate detrital Fe-grains
for source determinations when sand was present in the sample (Darby 2003,
2008). Adequate Fe-grains have been obtained from 5 gm of deep-sea muds
with ,2% sand (Darby 2008, 2014). However, 15 or more grams of bulk dry
sediment are usually needed with ,1–2% sand.
SAMPLE PREPARATION AND ANALYSIS
The 45–250 mm fraction is used here because grains smaller than this can be
difficult to identify under the microscope. Microscopic examination is needed
to confirm whether a grain contains multiple mineral phases (exsolution or
inclusions). After dispersal, wet sieving proceeds next with 250, 63, and 45 mm
sieves. Each fraction is dried and the highly magnetic grains, primarily
magnetite, are removed with a hand magnet so as to not clog the Frantz
magnetic separator. The 45 mm and 63 mm sieve fractions are run through the
Published Online: March 2015
Copyright
E
2015, SEPM (Society for Sedimentary Geology) 1527-1404/15/085-247/$03.00
Frantz magnetic separator using a side slope of 25uand 0.3 amp, but using a
steeper forward slope for the finer fraction so that both size fractions feed
through the Frantz at about the same rate.
The magnetic fractions from both of these size fractions along with the
hand magnetic separates are combined and mounted in one-inch-diameter
molds. Each circular mold is divided into quadrants using dividers made from
index cards cut about 0.5 cm width and that fit tightly into the circular molds.
We use either stainless steel or aluminum molds with six one-inch holes drilled
in a 25.5 cm 33.5 cm 31.5 cm plate. This drilled plate is then screwed onto
a solid plate that is about 0.6 cm thick with six screws. This bottom plate is
milled smooth to 3 mm, and a very thin coating of stopcock grease is applied
and wiped off with a lint-free tissue. This provides a base to which the grains
adhere, but if the grease is too thick, then the grains will not rest in the same
plane, measured in microns. The inside of the drilled holes in the upper plate
are liberally coated with grease to prevent the epoxy from adhering. A small
funnel is used to place the Fe-grains from the magnetic separation into each
quadrant of each circular plug. The location of each sample in each plug is
mapped to insure the integrity of the sample number. Epofix
TM
epoxy is
poured to just cover the dividers, and small 2-cm-diameter circular labels are
placed on top of the index card dividers with each sample number. More
epoxy is poured onto this label so that it is embedded within the sample plug.
They are polished first with 15 mm diamond paste using a lap wheel running at
about 200 rpm. Care must be taken not to grind through the smaller grains
but to expose as much of the larger grains as possible. During this initial
grinding the plugs are checked frequently under low-power magnification to
prevent over-grinding. Next, a 6 mm diamond paste and then a 3 mm paste are
used to complete the polishing.
The grains in each quadrant are photographed with a digital camera at 503
magnification. The Fe-grains are easily identified by their much higher
reflectance from non-iron grains under a reflected-light microscope using
10003magnification and immersion oil. As each grain is identified, it is
numbered on the photograph, producing a map for the probe operator. While
identification errors between some of the homogeneous Fe-oxide minerals
(ilmenite, titanomagnetite, magnetite, hematite, and chromite) are easily
corrected once the chemical compositions are determined, it is critical to
detect Fe-grains with more than one mineral phase, due either to exsolution
or to inclusions (Fig. 1). Ilmenite and hematite exsolutions are easily
identified by the large difference in reflectance between these minerals. Such
exsolutions are identifiable at sub-micron resolution with 10003. Reflected-
light microscopy along with grain chemistry are used to identify the mineral
type. Multi-phase minerals should be analyzed such that only one phase is
analyzed, preferably the phase with the greatest amount of chemical
variability. This would be the ilmenite or magnetite phase, whichever
dominates the grain. Hematite has less elemental substitution and thus is
less useful in determining a unique source (Table 1; Supplementary Materials,
Table A1). If the exsolution is too finely divided (,1–5 mm) to obtain an
analysis of just one phase, the grain should be skipped because of the
difficulty in matching grains with various proportions of two or more mineral
phases to source grains. Fewer than 30% of these grains can be matched.
Where the difference in reflectivity might be difficult to see on the microprobe
when programming an analysis, a small point is marked on the photomicro-
graph to guide the probe operator as to where the grain should be analyzed.
Finally, the plugs are thoroughly cleaned to remove immersion oil and
dried before coating with carbon for EPMA analysis. Analysis of all 14
elements takes less than five minutes per grain on a microprobe with five
spectrometers and large diffracting crystals such as the Cameca SX100 used
here with detectable limits in the tens to hundreds of parts per million
depending on the element (Table 1). Counting times were 20 s or 20,000
counts, whichever occurs first. Oxygen was analyzed using 120 s counting
time on an EDS system attached to the SX100 (see EPMA settings in
Supplementary Materials for more details). Precision is based on hundreds of
replicates for each of the Fe-oxide minerals (Supplementary Materials, Table
A2). We analyze as large an area in each homogeneous grain as possible in
order to measure the average composition of each grain. This accounts for
any variation within a grain. We use 1, 10, 20, and 40 mm beam sizes
depending on the size of the grain and whether an exsolved phase is present
and should be avoided (Fig. 1).
After the analyses are completed, the microscopic identifications are
checked against the composition for each grain and mineralogy is corrected as
needed. Exsolution, inclusions, and alteration observed during microscopic
examination are taken into account when determining mineralogy (see probe
coding criteria in Supplementary Materials).
MATCHING FE-OXIDE GRAINS TO SOURCES
Previous Matching Approach
Earlier uses of Fe-oxide grain matches to sources utilized discriminant
function analysis (DFA; Darby and Bischof 1996a, 1996b). This statistical
function provides a probability of group membership, but this in turn requires
that potential source samples be grouped by composition. This entails that
the potential source area samples be grouped by cluster analysis and repeated
tests and refinements of initial groups using DFA to test the uniqueness of
each grouping (Darby and Bischof 1996a; Darby et al. 2012). With large
numbers of potential source samples and areas, this can be a laborious
process, sometimes requiring more than five to ten iterations of each of
thousands of source composition groups.
While this procedure is proven to produce excellent grain matches, it takes
several days of running several hundred DFAs and then saving hundreds of
discriminant-function output files to search for the highest probabilities. In
addition, there is no easy way to determine if a different source might have
placed a close second. Here we present a much more efficient new matching
protocol that does not require grouping source samples by composition and
multiple DFAs.
New Matching Protocol
Instead of matching each Fe-grain from the deposit to compositional
groups of the same mineral in each source area (Fig. 2), the new protocol
matches each Fe-grain to every grain of the same mineral in the entire source-
area database. This source database consists of .38,000 grain analyses from
nearly 400 samples at 265 sample locations around the Arctic Ocean (Fig. 2).
Forty-five source areas were constructed based on both geographic location
of suspected sources for ice-rafted sediment and cluster analysis of
microscopic point counts of grain lithic types in the .250 mm size fraction
from these source samples (Bischof and Darby 1997, 1999). Sea ice can
entrain sediment in water depths of up to at least 50 m from any coastal area
around this ocean and transport it thousands of kilometers before melting.
Thus the entire circum-Arctic had to be sampled as densely as possible in
order to determine precise source locations. Most source areas in provenance
studies do not involve ice-rafting great distances and thus would require far
fewer samples.
The matching is done with a MATLABHroutine we wrote (Fig. 3; see
MATLAB routine in Supplementary Materials). The matching of Fe-oxide
grains with this database proceeds as follows: each element is compared with
that same element in all Fe-grains of the same mineral from the source
database (Fig. 3). Only grains with compositions within an accepted limit or
range of values for each element are matched and only when all 14 elements
F
IG
. 1.—Microphotograph of Fe-oxide grains and the four different beam sizes
used for different size grains. Magnetite (M), Ilmenite (Im), and hematite (H)
grains, some with exsolved phases are shown. Some of the exsolved hematite has
been leached or eroded, as seen by the dark blebs.
248 D.A. DARBY ET AL.
JSR
fall within the accepted range. For a match, the differences between each
element in the source grain and the grain to be matched must sum to less than
the sum of ranges for all elements (Table 1).
The logical statistical value to use for these ranges would be the average
standard deviation for each element on replicate analyses (61, 2, 3, or more
std. dev.). The replication error or range value for each element is based on
the average of 30–130 replications of diverse grains for each Fe-oxide mineral
type (Table 1; Supplementary Materials, Table A2). For these replications,
we chose three or more Fe-oxide grains from each of the Fe-oxide minerals
lacking exsolved phases or inclusions (e.g., fresh ilmenite, slightly altered
ilmenite, magnetite, titanomagnetite, hematite, and chromite). More grains
were used for minerals having greater compositional variations. For example,
130 replicated spot analyses in 20 grains of magnetite were used (Table 1).
Each grain chosen had elevated levels of as many of the 14 elements as
possible and was analyzed in five or ten different spots, each less than five
microns in size. The low variance in these replicates indicates that there were
no exsolution features or inclusions in these grains. Different minerals contain
contrasting amounts of even the most common elements (Fe, Ti, and O,
Table 1).
One option for determining the range of values for each element for a
match is to use the average variance for several replications (Table 1). The
five or ten analyses in a selected grain were averaged and a standard deviation
was calculated for each element in each grain. These variances for each grain
were averaged for all the grains of that mineral to obtain the overall average
standard deviation for each element in that mineral type (Table 1). The
variance or standard deviation is affected only by the analytical precision and
the variability within each grain. Another option is to use the average
variance for each element for all the replicates in all of the different grains of
that mineral used for replicates. In this instance, the variance is a combination
of analytical error, variability within each grain, and variation among all
grains of the same mineral. Use of this method results in larger standard
deviations, especially for minerals like titanomagnetite that have large
amounts of substitution (Table 1).
Accuracy of the Matching Protocol
The accuracy of the matching is tested in two ways, first by testing grains
with known sources and second by comparing with the previous matching
protocol that used DFA to match grains to source groups. The objective of
the new protocol is to speed the matching, decrease erroneous matches,
include matches to other grains of similar composition, avoid the need for
labor-intensive grouping and testing of source grains, and maximize the
T
ABLE
1.—One standard deviation on average replicate values of Fe oxide grains where either five or ten different spots we re reanalyzed on each grain.
Number of
Replicates Ti Fe Mn Mg Si Al Cr Zn V Ca Nb Ta Ni O
Sum of Std
Dev
Fresh ilmenite 70 1 std dev - avg of each grain 0.25 0.05 0.01 0.01 0.01 0.01 0.01 0.01 0.01 0.03 0.02 0.01 0.52 1.04 1.99
1 std dev - avg of all grains 0.74 0.92 1.21 0.09 0.04 0.01 0.01 0.02 0.03 0.01 0.06 0.03 0.01 0.88 4.05
Altered ilmenite 30 1 std dev - avg of each grain 1.09 1.98 0.68 0.04 0.04 0.01 0.01 0.02 0.02 0.01 0.06 0.03 0.01 0.97 4.98
1 std dev - avg of all grains 1.32 3.21 2.34 0.11 0.07 0.02 0.01 0.03 0.05 0.01 0.15 0.04 0.01 1.67 9.05
Hematite 30 1 std dev - avg of each grain 0.02 0.52 0.01 0.01 0.01 0.03 0.02 0.02 0.04 0.01 0.02 0.03 0.01 0.70 1.43
1 std dev - avg of all grains 0.03 1.05 0.01 0.01 0.03 0.06 0.04 0.02 0.11 0.01 0.02 0.03 0.02 1.83 3.29
Magnetite 130 1 std dev - avg of each grain 0.02 0.43 0.01 0.01 0.03 0.03 0.01 0.01 0.01 0.02 0.02 0.02 0.04 0.67 1.33
1 std dev - avg of all grains 0.06 3.12 0.06 0.64 1.08 0.13 0.07 0.02 0.22 0.05 0.02 0.02 0.75 5.87 12.12
Chromite 40 1 std dev - avg of each grain 0.09 1.34 0.07 0.27 0.01 0.22 1.13 0.03 0.02 0.01 0.03 0.02 0.01 1.04 4.30
1 std dev - avg of all grains 0.22 4.09 0.24 0.93 0.01 0.57 3.07 0.08 0.05 0.01 0.03 0.03 0.02 1.35 10.71
Titano-magnetite 90 1 std dev - avg of each grain 0.51 1.35 0.07 0.09 0.34 0.12 0.05 0.02 0.02 0.18 0.01 0.01 0.01 0.68 3.46
1 std dev - avg of all grains 3.54 4.47 0.36 0.97 1.03 0.42 0.99 0.05 0.19 1.06 0.01 0.01 0.02 1.45 14.57
F
IG
. 2.—Location of source samples and
source areas numbered 1–45. Source areas were
designated initially by geographic location and
unique groupings of lithic grains .250
m
m
(Darby and Bischof 1996a, 1996b). Core MD99-
2263, on the western Icelandic shelf at Denmark
Strait, is located in 235 m water depth.
METHOD FOR PRECISE SOURCE DETERMINATION 249
JSR
percentage of grains matched to a source area with acceptable error
margins.
To test the accuracy or error of the match protocol we used the large
(,38,000 grain) source database for the circum-Arctic (Supplementary
Materials; Table A1; Fig. 2). Each grain in this database is matched to every
other grain of the same mineral type. Using the average one standard
deviation on each element results in a very low match rate of only 0.1%, which
increases only to 76% with six standard deviations (Table 2). The low match
rate indicates that the EPMA precision is adequate to distinguish thousands
of grains. The small average standard deviations on each of the 14 elements
suggest that these Fe-grains are fairly homogeneous.
We tested one to six standard deviations around each element to find the
optimum number of matches with acceptable errors in Fe-grain matches of
the same mineral in the source database (Table 2; Supplementary Materials,
Tables A3, A4). Using a 2swindow for each element resulted in only 49% of
the grains matching to another grain in the source data set of the same
mineral type, a result no better than using DFA. Using the variance of all
replicates of a specific mineral resulted in 78% of grains matching to source
grains using a 2swindow and 94% match using 6s.
The key is to match as many grains as possible to a source area and still
have a high degree of accuracy (i.e., matches to Fe-oxide grains from the
same source area). As the window about each element is increased from 2 to
6s, the percentage of grains matched to a source increases but the average
percentage of grains incorrectly matched to each source area increases
(Table 2). Regardless of the size of the window or range in variance for each
element used in the match protocol, the average error of mismatches in each
source area was always 2% or less of those grains that were successfully
matched (Table 2). The average maximum percentage of mismatched grains
F
IG
. 3.—Flow chart of the match protocol in MATLABH.
250 D.A. DARBY ET AL.
JSR
to source areas increases rapidly with larger ranges but levels off around
11%.
With the DFA it is difficult to identify source areas that might place a close
second or third. To account for these non-unique match situations, we allow
each grain more than just the best source-grain match. Thus a grain could
match to more than one source area and we prorate the source determination
depending on the closeness of elemental fit between the source grains and the
grain being matched. In order to do this, the difference between the grain
being matched and each matched source grain for each element is calculated.
These differences are summed for all 14 elements and the closeness of match is
evaluated using this sum. The source grains are weighted proportionately by
summing the inverse sums and scaling to one. Thus, if only one grain matched
it would have a weighted value of one. If two source grains are matched and
one source-area grain had a summed difference of 3.0% for all 14 elements
and the second grain a sum of 4.0%, then the inverses would be 0.33 and 0.25,
respectively. Dividing these values by their sum (0.58), we find the
proportional match value for each source, 0.57 and 0.43, respectively, which
sum to 1.0. Thus, the grain being matched would be assigned to the first
source area at a value of 0.57 and the second at 0.43. If both source grains are
from the same area, then these values are summed and the source would be
assigned only to that one source. Once all grains are matched, the source-area
values are summed to find the weighted proportion to each possible source.
Only grains with sum deviations for all elements less than the sum of the
standard deviations from the replicate analyses of each element are prorated
in this way. Source-area grains that have summed deviations greater than this
are not matched (see Table 1).
The range that provides the highest percentage of matches with the least
error is provided by 2sof all replicates. The average maximum incorrect
match for each source area using this range is 10% (Table 2). Part of the
reason that this maximum error is even this large is that there are several
sources composed of samples from shelf areas where sea ice could deliver
grains from distant sources (Fig. 2), resulting in matches to these distant
sources (Bischof and Darby 1999).
We tested 1sfor magnetite grains while using 2sfor all other minerals
because the 2sranges for magnetite were so large (Supplementary Materials,
Table A2). This change made little difference in the match rate or error
percentages (Table 2), which reflects the robustness of the Fe-grain
fingerprinting method and the procedure of matching each grain propor-
tionally to the source(s) with the closest overall composition.
The second way in which we tested the new match protocol is by comparing
the matches with those using the DFA for samples from the same core. Forty-
three samples from core JPC16 (Fig. 2) were reanalyzed using the Cameca
SX100 that were previously analyzed on an older ETEC autoprobe to remove
any possibility of machine differences. Not all of the same grains could be
analyzed because in re-polishing the samples for analysis some of the original
grains were lost. Despite this, similar numbers of grains were analyzed in each
sample and the average difference in the two match protocols is one grain per
source area with an average maximum difference of six grains for any source
area. The match percentage is not very different for the two match protocols, 50
and 55.5%, respectively for the old and new methods for the samples in this core.
The average difference in the number of Fe-grains matched to each source
in another core (ACEX, Fig. 2) using 16 samples analyzed by the SX100 but
T
ABLE
2.—Test of source area dataset matches using diferent element ranges based on standard deviations (SD). Each of ,38,000 Fe-oxide grains in this
dataset is matched to all other grains of the same mineral type. Different matching criteria are used for the range of each element and the total difference of
all summed elements. Criteria are from replicate values in Table 1.
6Range Criteria for
Each Element % Matched
AVG % of All Grains
Correctly Matched
in Each Source Area
AVG % of All Grains
Correctly Matched to Its Source
and Nearby Sources (300 km)
AVG % of Those
Matched Grains
Incorrectly Matched
Max Avg %
Incorrectly
Matched
Using the average variance for each replicate grain for each mineral
1 SD 0.1 100 100 0.02 0.02
2 SD 49 59 97 0.1 1
4SD 54 58 62 1.01 7
6SD 76 36 50 1.45 9
Using the average variance of all replicates for each mineral
1 SD 31 75 78 0.6 4
2 SD 78 33 41 1.52 10
2 SD & Mag.1 SD 75 36 44 1.46 9
4SD 91 17 25 1.9 11
4 SD & Mag.2 SD 90 18 26 1.9 11
4 SD & Mag.3 SD 91 17 26 1.9 11
6SD 94 12 21 2 11
T
ABLE
3.— Comparison of matching methods using different element ranges based on standard deviations (SD) using sediment cores. Analyses were
preformed on the same instrument, a Cameca SX100 EPMA.
ACEX Core Intervals
90–100 Grains Analyzed % Matched
Avg Number Differ for
Each Core Depth
Avg Max for
Each Core Depth
Avg Num Grains Matched
for Each Core Depth
2 SD each replicate 4.4 2 14 3.9
4 SD each replicate 70.0 2 10 62.8
2 SD all replicates 87.6 2 10 78.8
4 SD all replicates 96.9 21087.3
Disc. Fn Matches 73.7 73.1
JPC16 core intervals
2 SD each replicate 0.5 1 6 0.4
4 SD each replicate 17.9 1 6 11.6
2 SD all replicates 55.5 1 9 39.5
4 SD all replicates 83.7 1860.1
Disc. Fn Matches 50.0 35.9
METHOD FOR PRECISE SOURCE DETERMINATION 251
JSR
matched by the two different methods is two grains with an average
maximum difference of 10 grains for any source area. In this case, the match
percentages were much higher, 87.6 for the new match protocol using 2s
compared to 73.7% for DFA (Table 3). The reason for the match
percentage difference in the two cores (55.5% vs. 87.6%) is unknown, but
it probably has to do to with the location of the JPC16 core close to the
Alaskan shelf and the fact that we may not have sampled all of the potential
local sources.
Based on this limited test between the DFA method and the new protocol,
a case can be made for using 4sinstead of two (Table 3). The match
percentages are much higher using 4son all replicates (83.7% vs. 55.5% in
JPC16 and 96.9% vs. 87.6% in the ACEX core samples). However, the
significant decrease in the percentage of correctly matched grains to each
source from 33 to 17% for two vs. four standard deviations based on
matching grains in the circum-Arctic data set (Table 2) would favor the 2s
range. In cases where the 2srange does not result in sufficient numbers of
grains matched to a source, the range could be increased to improve the
match percentage without a significant increase in the average match error
(1.5% vs. 1.9% for the 2 vs. 4srange, respectively; Table 2).
Example of the New Match Protocol
To illustrate this new protocol and the usefulness of the Fe-oxide chemical
fingerprinting method, a core off western Iceland (Fig. 2) is analyzed in order
to determine the input from large glaciers on east and southeast Greenland (SA
42 and 43, respectively, Fig. 2) versus very distant sources in the Arctic Ocean
(Fig. 4; see Andrews et al. 2009 for another example of this new protocol). The
most important of these sources in order are Banks Island (SA8), Svalbard
(SA33), and northern Ellesmere Island (SA4), all with maximum grain matches
of $5 grains. Fe-oxide grains from Banks Island drifted more than 6,000 km
to reach this core site. As expected, the dominant sources of ice-rafted Fe-oxide
grains in this core are the east and southeast Greenland glaciers.
F
IG
. 4.—Fe-oxide grain matches in core MD99-2263 at Denmark Strait showing small but significant input of ice-rafted grains from the Arctic Ocean as well as a large
input from E and SE Greenland glaciers. The dashed line near the bottom is the average error of mismatches (1.5% of matches for each sample) resulting in an average
error of 0.9 60.1 grains. The maximum number of grains from any one Arctic Ocean source (SA1-41 in Fig. 2) is always significant, so that every sample contains some
Arctic input. This input varies with larger peaks in the last millennium and is not in phase with Greenland iceberg input.
252 D.A. DARBY ET AL.
JSR
CONCLUSIONS
The matching or source-determination protocol presented here provides a
relatively fast and efficient method for determining the sources of detrital
sediment where the proportion of multiple sources can be ascertained. The
statistical accuracy of this method is high (incorrect matches average less than
2%), but it cannot be compared with other provenance techniques for lack of
such statistical tests on other methods. Fe-grain fingerprinting can be used on
nearly any detrital sediment and can distinguish multiple sources in a complex
mixture of sources. The Fe-grains used are not only relatively abundant in
detrital sediment but also are stable over millions of years and easily extracted
for analysis. The new match protocol does not require labor-intensive grouping
of Fe-grain composition like DFA, so that all that is needed is analysis of about
100 grains in samples from potential sources. The source determinations are
statistically valid and easily tested, unlike many other provenance techniques
where very few grains are typically analyzed due to the expense of the analysis
or the rarity of the required detrital grains in most sedimentdeposits or the time
required extracting these mineral grains. Fe-grain fingerprinting provides good
statistical matches to precise sources (Table 2) such as individual rivers or
source areas consisting of just one sample (Fig. 2). For example, significant
numbers of Fe-grains were matched to source area 37 (Fig. 2, Vilkitsky Strait)
that is represented by only one sample. Several other source areas are
represented by only 1–3 samples, and they too have had significant numbers of
Fe-grains matched to them from some samples.
ACKNOWLEDGMENTS
This paper is dedicated to the memory of Michele L. Darby (17 August
1949–5 February 2015). Wife, mother, educator, and scholar, she inspired
thousands around the world with her devotion to education in her field of
dental hygiene, her warmth of character, and her generosity of spirit. NSF
provided the funding for this research. Helpful reviews were provided by A.
Basu, R. Ingersoll, S. McLennan, and A. Morton.
SUPPLEMENTAL MATERIAL
Supplemental files are available from JSR’s Data Archive: http://sepm.org/
pages.aspx?pageid5229. An earlier version of the circum-Arctic data used for
source determination in this paper is archived at the ACADIS data archive:
https://www.aoncadis.org/dataset/provenance.html.
REFERENCES
A
NDREWS
, J.T., D
ARBY
, D.A., E
BERL
, D., J
ENNINGS
, A.E., M
OROS
,M.,
AND
O
GILVIE
,A.,
2009, A robust, multi-site Holocene history of drift-ice off northern Iceland:
implications for North Atlantic climate: The Holocene, v. 19, p. 71–77.
B
ASU
,A.,
AND
M
OLINAROLI
, E., 1989, Provenance characteristics of detrital opaque Fe-
Ti oxide minerals: Journal of Sedimentary Petrology, v. 59, p. 922–934.
B
ISCHOF
, J.A.,
AND
D
ARBY
, D.A., 1997, Mid to Late Pleistocene ice drift in the western
Arctic Ocean: evidence for a different circulation in the past: Science, v. 277, p. 74–78.
B
ISCHOF
,J.,
AND
D
ARBY
, D., 1999, Quaternary ice transport in the Canadian Arctic and
extent of late Wisconsinan glaciation in the Queen Elizabeth Islands: Canadian
Journal of Earth Sciences, v. 36, p. 2007–2022
B
LATT
, H., M
IDDLETON
,G.,
AND
M
URRAY
, R., 1972, Origin of Sedimentary Rocks:
Englewood Cliffs, N.J., Prentice-Hall, 634 p.
C
ORNELL
, R.M.,
AND
S
CHWERTMANN
, U., 2003, The Iron Oxides: Structure, Properties,
Reactions, Occurrences and Uses, Second Edition: New York, Wiley, 664 p.
C
RAIG
, J.A.,
AND
V
AUGHAN
, D.J., 1994, Ore Microscopy and Ore Petrography, Second
Edition: Mineralogical Society of America, 434 p.
D
ARBY
, D.A., 1984, Trace elements in ilmenite: a way to discriminate provenance or age
in coastal sands: Geological Society of America, Bulletin, v. 95, p. 1208–1218.
D
ARBY
, D.A., 1990, Evidence for the Hudson River as the dominant source of sand on
the U.S. Atlantic Shelf: Nature, v. 346, p. 828–831.
D
ARBY
, D.A., 1998, Mysterious iron-nickel-zinc spherules: Canadian Journal of Earth
Sciences, v. 35, p. 23–39.
D
ARBY
, D.A., 2003, Sources of sediment found in sea ice from the western Arctic Ocean,
new insights into processes of entrainment and drift patterns: Journal of Geophysical
Research, v. 108, no. C8, Paper 3257, p. 1–10.
D
ARBY
, D.A., 2008, The Arctic perennial ice cover over the last 14 million years:
Paleoceanography, v. 23, p. 1–9.
D
ARBY
, D.A., 2014, Ephemeral formation of perennial sea ice in the Arctic Ocean during
the middle Eocene: Nature Geoscience, v. 7, p. 210–213.
D
ARBY
, D.A.,
AND
B
ISCHOF
, J.F., 1996a, A statistical approach to source determination
of lithic and Fe-oxide grains: an example from the Alpha Ridge, Arctic Ocean:
Journal of Sedimentary Research, v. 66, p. 599–607.
D
ARBY
, D.A.,
AND
B
ISCHOF
, J.F., 1996b, Sources of ice-rafted detritus and iceberg tracks
in the Arctic Ocean, in Tucker, W., and Cate, D., eds., The 1994 Arctic Ocean Section:
The First Major Scientific Crossing of the Arctic Ocean: Hanover, New Hampshire,
U.S. Army Cold Regions Research and Engineering Lab, p. 89–91.
D
ARBY
, D.A.,
AND
B
ISCHOF
, J.F., 1998, Evidence for the Chukchi Sea as an important
source of sea-ice entrainment and the role of sea-ice in transporting fine-grained
sediment from one shelf to another in the Arctic: EOS, v. 79, no. 47, p. F438.
D
ARBY
, D.A.,
AND
E
VANS
, A.E., J
R
., 1992, Provenance of Quaternary beach deposits,
Virginia and North Carolina, in Fletcher, C., and Wehmiller, J., eds., Quaternary
Coasts of the United States: Marine and Lacustrine Systems: SEPM, Special
Publication 48, p. 113–119.
D
ARBY
, D.A.,
AND
T
SANG
, Y.W., 1987, The variation in ilmenite element composition
within and among drainage basins: implications for provenance: Journal of
Sedimentary Petrology, v. 57, p. 831–838.
D
ARBY
, D.A.,
AND
Z
IMMERMAN
, P., 2008, Ice-rafted detritus events in the Arctic during
the last glacial interval and the timing of the Innuitian and Laurentide Ice Sheet
calving events: Polar Research, v. 27, p. 114–127.
D
ARBY
, D., B
ISCHOF
, J., C
UTTER
,G.,
DE
V
ERNAL
, A., H
ILLAIRE
-M
ARCEL
, C., D
WYER
,G.,
M
C
M
ANUS
, J., O
STERMAN
, L., P
OLYAK
,L.,
AND
P
OORE
, R., 2001, New record of
pronounced changes in Arctic Ocean circulation and climate: EOS, v. 82, p. 603–607.
D
ARBY
, D.A., B
ISCHOF
, J., S
PIELHAGEN
, R., M
ARSHALL
,S.,
AND
H
ERMAN
, S., 2002, Arctic
ice export events and their potential impact on global climate during the late
Pleistocene: Paleoceanography, v. 17, p. 15.1–15.17.
D
ARBY
, D.A., M
YERS
, W.B., J
AKOBSSON
,M.,
AND
R
IGOR
, I., 2011, Modern dirty sea ice
characteristics and sources: the role of anchor ice: Journal of Geophysical Research,
v. 116, no. C09008, p. 1–18.
D
ARBY
, D.A., O
RTIZ
, J.D., G
ROSCH
,C.,
AND
L
UND
, S., 2012, 1,500 year cycle in the
Arctic Oscillation identified in Holocene Arctic sea-ice drift: Nature Geoscience, v. 5,
p. 897–900.
D
IMANCHE
,F.,
AND
B
ARTHOLOME
, P., 1976, The alteration of ilmenite in sediments:
Minerals Science Engineering, v. 8, p. 187–201.
F
ORCE
, E.R., 1991, Geology of titanium-mineral deposits, Chapter 1: Geological Society
of America, Special Paper 259, p. 1–57.
G
ASPAR
, J.C.,
AND
W
YLLE
, P.J., 1983, Ilmenite (high Mg, Mn, Nb) in the carbonates
from Jacupiranea Complex, Brazil: American Mineralogist, v. 68, p. 960–971.
G
RIGSBY
, J.D., 1992, Chemical fingerprinting in detrital ilmenite: a viable alternative in
provenance research?: Journal of Sedimentary Petrology, v. 62, p. 331–337.
H
AGGERTY
, S.E., 1976, Opaque mineral oxides in terrestrial igneous rocks, in Runble,
D., ed., Oxide Minerals: Reviews in Mineralogy, v. 3, p. 101–300.
L
EWIS
, J.F., 1970, Chemical composition and physical properties of magnetite of the
ejected plutonic blocks of the Soufriere Volcano, St. Vincent, West Indies: The
American Mineralogist, v. 55, p. 793–807.
M
ANGE
, M.A.,
AND
W
RIGHT
,D.T.,
EDITORS
, 2007, Heavy Minerals in Use: Amsterdam,
Elsevier, Developments in Sedimentology, no. 58, 1283 p.
M
ORTON
, A., H
ALLSWORTH
,C.,
AND
C
HALTON
, B., 2004, Garnet compositions in Scottish
and Norwegian basement terrains: a framework for interpretation of North Sea
provenance: Marine and Petroleum Geology, v. 21, p. 393–410.
R
ASBURY
, E.T., H
EMMING
, S.R.,
AND
R
IGGS
, N.R.,
EDITORS
, 2012, Mineralogical and
Geochemical Approaches to Provenance: The Geological Society of America, Special
Paper 487, 194 p.
S
UAVET
, C., G
ATTACCECA
, J., R
OCHETTE
, P., P
ERCHIAZZI
, N., F
OLCO
, L., D
UPRAT
,J.,
AND
H
ARVEY
, R.P., 2009, Magnetic properties of micrometeorites: Journal of Geophysical
Research, v. 114, no. B04102.
VAN
A
NDEL
, T.H., 1959, Reflections on the interpretation of heavy mineral analyses:
Journal of Sedimentary Research, v. 29, p. 153–163.
W
AYCHUNAS
, G.A., 1991, Crystal chemistry of oxides and oxyhydroxides: Reviews in
Mineralogy, Oxide Minerals, Mineralogical Society of America, v. 24, p. 11–68.
Received 25 August 2014; accepted 16 November 2014.
METHOD FOR PRECISE SOURCE DETERMINATION 253
JSR
