A global compilation of nearly sixty measurement studies is used to evaluate two methods of simulating the mineral composition of dust aerosols in an Earth system model. Both methods are based upon a Mean Mineralogical Table (MMT) that relates the soil mineral fractions to a global atlas of arid soil type. The Soil Mineral Fraction (SMF) method assumes that the aerosol mineral fractions match the fractions of the soil. The MMT is based upon soil measurements after wet sieving, a process that destroys aggregates of soil particles that would have been emitted from the original, undisturbed soil. The second method approximately reconstructs the emitted aggregates. This model is referred to as the Aerosol Mineral Fraction (AMF) method because the mineral fractions of the aerosols differ from those of the wet-sieved parent soil, partly due to reaggregation. The AMF method remedies some of the deficiencies of the SMF method in comparison to observations. Only the AMF method exhibits phyllosilicate mass at silt sizes, where they are abundant according to observations. In addition, the AMF quartz fraction of silt particles is in better agreement with measured values, in contrast to the overestimated SMF fraction. Measurements at distinct clay and silt particle sizes are shown to be more useful for evaluation of the models, in contrast to the sum over all particles sizes that is susceptible to compensating errors, as illustrated by the SMF experiment. Model errors suggest that allocation of the emitted silt fraction of each mineral into the corresponding transported size categories is an important remaining source of uncertainty. Evaluation of both models and the MMT is hindered by the limited number of size-resolved measurements of mineral content that sparsely sample aerosols from the major dust sources. The importance of climate processes dependent upon aerosol mineral composition shows the need for global and routine mineral measurements.

The effect of soil dust aerosols upon climate is dependent upon the
particle mineral composition

A second challenge is how to treat particles that are combinations of
different minerals. For example, iron oxides are often observed as
small impurities attached to particles comprised predominantly of
other minerals

Finally, refinement of models is challenged by limited global measurements of size-resolved aerosol composition. Many of the available measurements are from field campaigns or ship cruises of limited duration, while changes in the sampling and analysis methods through time have contributed additional uncertainty.

We address the first two challenges in a companion paper

We also propose a method for mixing minerals with small impurities of
iron oxides, which we call “accretions”. In our model, iron oxides
can travel either in pure crystalline form or as accretions
internally mixed with other minerals. The distribution of the two
forms of iron oxide is based on the degree of weathering that creates
iron oxides in the soil

In this article, we compare our calculation of aerosol mineral
content to a new global compilation of observations from almost sixty
citations. In Sect. 2, we summarize our new modeling approach and the
simulations performed with the NASA Goddard Institute for Space
Studies (GISS) Earth System ModelE2, whose details can be found in
the companion article

Simulations are performed with the CMIP5 version of the NASA GISS Earth
System ModelE2

Two simulations are compared to our compilation of observations. The
control or “baseline” simulation assumes that the emitted mineral
fractions are identical to those of the wet-sieved parent soil; this
calculation is referred to as the Soil Mineral Fraction (SMF) method.
The soil (and thus the emitted) mineral fractions are calculated by
combining the Mean Mineralogical Table

The MMT provides the fractional abundance for eight minerals within
the clay and silt-size ranges of the soil as a function of arid soil
type. For the clay-size range (whose diameters are less than

The mineral fractions provided by the MMT for each size category are combined with the mass fraction of each size category provided by the soil texture atlas. This gives the size-resolved mineral fractions of the wet-sieved soil at each location.

After emission, the minerals are transported within five size classes
with diameters extending between 0.1 and

Dust at Tinfou is measured after transport from the source, when the
largest particles are removed preferentially by gravitational
settling.

The allocation of silt-sized emission within the individual size categories
transported by ModelE2 is empirical and based upon measurements at only a
single location. It is difficult to test the validity of this allocation at
other locations, given the paucity of size-resolved measurements of mineral
fractions. At diameters above roughly 20

Our second simulation is motivated by measurements showing significant
differences between the size-resolved mineral fractions of wet-sieved soils
and aerosol concentration. This simulation is referred to as the Aerosol
Mineral Fraction (AMF) method to emphasize the difference between the aerosol
and soil mineral fractions (in contrast to the SMF where these fractions are
assumed to be identical). This difference results because wet sieving is more
destructive of aggregates of soil particles than mobilization of the
original, undispersed soil, where many of the aerosols are comprised of
aggregates that resist complete disintegration during emission. Brittle
fragmentation theory provides a physically based method for reconstructing
the emitted size distribution from the distribution measured after wet
sieving

Conversely, the MMT provides the fraction of feldspar and gypsum only
at silt sizes, even though aerosol measurements show that these
minerals are present at both clay and silt sizes

To apportion the emitted silt fraction of the AMF simulation into the
ModelE2 transport categories, we combine the size distribution derived
from brittle fragmentation theory (that is valid for diameters below
roughly 20

Finally, for the AMF experiment, we allow iron oxides to be emitted
not only in their pure, crystalline form, but additionally as
impurities mixed with other minerals. These mixtures are important
for transporting iron far from its source, because pure iron oxides
are more dense and vulnerable to gravitational removal than most other
minerals that contain small inclusions or accretions of iron oxides.
We assume that the partitioning of iron oxides into mixtures and pure
crystalline forms depends upon the soil fraction of iron oxides
compared to the other minerals (as given by the MMT, including our
extension to clay sizes). Soils enriched in iron oxides are assumed
to be highly weathered, with a greater abundance of the pure,
crystalline form

The dust aerosol module described by

Dust sources are prescribed within topographic depressions

Dust removal results from wet and dry deposition. The latter includes
gravitational settling and turbulent deposition in the surface layer

Wet deposition has been updated since its description in

Measurements show that physical and chemical properties of aerosols
evolve along their trajectory

We also defer calculation of radiative forcing as a function of the aerosol mineral composition. As a result, radiative feedbacks between the mineral fractions and climate are disabled.

Both the SMF and AMF simulations are performed with ModelE2 at a resolution
of

List of literature references for mineral fraction measurements (predicted with ModelE2: M – mica/illite/muscovite, K – kaolinite, S – smectite, C – carbonates, Q – quartz, F – feldspar, I – iron oxides, G – gypsum; other minerals not predicted: O) with specific information about months of measurements with size range, geographical coordinates, and time range of measurements.

Continued.

Continued.

Locations of measured mineral fractions compiled from the literature
used for the evaluation of the simulations. References with
geographical coordinates in the legend provide measurements only for
this single location; otherwise, references provide measurements for
multiple locations. See Table

We compiled measurements of mineral fractions of dust aerosols from
almost sixty studies published between the 1960s and the present day
that are described in Table

Methods to determine the mineral composition of dust aerosols have
varied over time, and the measurements in our compilation that are
based on various instruments and analytical methods contain different
biases and uncertainties. Systematic studies of the mineral
composition of atmospheric soil dust started in the 1960s, beginning
with

Since the 1990s, airborne dust has been more commonly sampled with
other instruments, like high-volume air samplers

The relative mass fractions of the collected minerals are often derived from
X-ray diffraction (XRD) spectra

XRD analysis is most effective for minerals with a regular crystal structure
whose spectral peaks are well defined. However, certain minerals like
phyllosilicates consist of varying amounts of amorphous material whose mass
is difficult to quantify using XRD

The composition of airborne particles is increasingly studied by
scanning electron microscope (SEM) images of individual particles
along with statistical cluster analysis of elemental composition

All the observations used for our evaluation are based on
measurements of the mineral fractions of dust aerosols at the
surface. A few studies also provide aircraft measurements

Because of the difficulty of comparing the uncertainty of different measurement methods, we weight all observations equally. As prognostic models of mineral composition become more common, we hope that mineral identification within aerosol samples becomes more uniform and routine.

A challenge for model evaluation is the difference in record length between
climate model output and the mineral observations. Deposition is measured
over periods as short as a week. Measurements of surface concentration are
based mostly on daily sampling, with reported values derived from a few days.
In contrast, the output from our model simulations consists of a continuous
stream of data, from which monthly averages are calculated. Note that even
though the model output could be archived at higher frequencies, e.g., every
model day, a large discrepancy between the small sample sizes of many of the
measurements and large samples from the model simulations would persist. The
mineral fractions that we use for evaluation reflect the composition of the
soil at the source region. These fractions are probably more consistent than
the absolute concentration of the separate minerals used to form this ratio,
at least in those remote regions where a single source dominates the supply.
Thus, measurements of mineral fractions from a few days may be representative
of the entire month. Closer to a source, the mineral fractions may be more
variable, with episodic increases of quartz and other minerals that are
abundant at large diameters during dust storms

For each reference providing measurements, we calculate a time average that can be compared to the model output. In some cases, we estimate a monthly average using daily measurements that are available for only a subset of the month. Our simulations cover only the 9 years between 2002 and 2010, but some of the measurements date back to the 1960s. Our evaluation assumes that multi-decadal variability in the mineral fractions of dust aerosols at individual locations is small compared to the fractions themselves. A more thorough discussion of the sampling uncertainty in our comparison between the measurements and model is provided in Appendix A.

We simulate only eight minerals in our model. However, measurements may
include additional minerals that are not simulated. Other measurements may
not include all of the simulated minerals. (For example,

To account for different size ranges of the model and measurements, we
interpolate the mass fractions from the model size bins to the size
range of the measurements. For measurements of total suspended
particles (TSP), we compare to the sum over the entire model size
range. Since this range extends only to

We compare the measured and simulated mineral fractions and ratios using
scatterplots. We calculate the normalized bias (nBias) and normalized root
mean squared error (nRMSE). Normalization was done by dividing the statistic
by the average of the observed values used in each scatterplot. The number of
paired data points (

Our evaluation compares measurements from a specific location to the value at
the corresponding grid box. In the case of ship cruises, we use the average
along the cruise trajectory within each ocean, forming a model average with
the corresponding sequence of grid boxes. Our comparison assumes that the
grid size of the model is sufficient to resolve spatial variations of the
measurements. This is not always the case, particularly near dust sources
that are often geographically isolated, resulting in strong spatial contrasts
of concentration

In a companion paper

Below, we extend the evaluation of both methods to the global scale. We calculate mineral fractions that are the ratio of the mass of each mineral to the sum over all minerals. Alternatively, we consider the ratio of specific mineral pairs. The mineral mass is derived from surface concentration or deposition, depending upon the measured quantity.

Annual cycle of illite plus smectite and kaolinite
fractions for diameters less than

Same as Fig.

Only a few locations have measurements at multiple times throughout the year, and these are generally insufficient to resolve the seasonal cycle. We use these measurements for comparison to the model that at some locations exhibits a seasonal shift in the predominant mineral.

Figure

Over the Pacific, both the SMF and the AMF experiments show similar illite–smectite and kaolinite fractions at clay sizes that are consistent with the observations. The slightly smaller AMF fraction of phyllosilicates results from the addition of feldspar and gypsum at clay sizes that comes at the expense of the phyllosilicate fraction. (This difference between the AMF and SMF treatments of phyllosilicates is obscured in the Barbados measurements because feldspar and gypsum are not measured and are thus excluded from our reconstruction of the total dust mass at clay sizes.) At silt sizes, the simulated AMF fraction of phyllosilicates that is observed at the Pacific locations is entirely absent in the SMF experiment, highlighting the importance of reconstructing the emitted phyllosilicate mass comprised of soil aggregates that are almost totally disintegrated during wet sieving of the soil samples. There is the suggestion that the kaolinite fraction is overestimated by the model at both clay and silt sizes, a discrepancy that is found at other locations, as will be discussed below.

Figure

For feldspar, the AMF method reproduces the clay-size fraction of most measurements, in contrast to the SMF experiment, which omits feldspar at this size. At silt diameters, both experiments are consistent with the measurements, owing in part to their large uncertainty.

Scatterplot of mineral fractions of illite, kaolinite, the
sum of illite and smectite, all phyllosilicates and quartz for
silt particles (whose diameters are greater than

Same as Fig.

Same as Fig.

We summarize the model performance by comparison to a global distribution of
measurements at silt and clay diameters, respectively
(Figs.

Figure

At silt diameters, the SMF method systematically overestimates the
observed quartz fraction while entirely excluding the phyllosilicates
(Fig.

Even with reaggregation, the AMF method tends to underestimate illite at silt sizes, while overestimating kaolinite and smectite (the latter not shown). These errors could result from the mineral fractions prescribed by the MMT at silt sizes, but also from the MMT clay fractions due to reaggregation. Combinations of illite with the other phyllosilicates show better agreement.

Figure

All three experiments show good agreement of the quartz fraction at
clay sizes (Fig.

Same as Fig.

Bulk measurements of mineral composition represent sums over all
particle sizes, and are plentiful compared to measurements within
individual size categories. Both the SMF and AMF methods produce
similar bulk fractions of phyllosilicates
(Fig.

With the exception of source regions and their vicinity, the AMF and SMF
methods produce bulk fractions of both total phyllosilicates and quartz that
are in good agreement with the measured values
(Figs.

This compensation is disabled in the AMF experiment with

Conversely, fractional emission at clay sizes for

All the experiments exhibit negative biases for their fractions of
carbonates, gypsum, and iron oxide (Fig.

Measurements over the Arabian Peninsula

Measured vs. simulated mineral ratios with respect to
quartz for the SMF, AMF and AMF (

The mineral fractions with respect to total dust that are analyzed in the previous section are unaffected by model errors in global emission. For consistency, we constructed the total dust mass using only minerals that are common to both the model and the specific measurement study. However, this construction introduces errors where measurements of total dust include minerals that are not reported. By considering ratios of specifc pairs of minerals, we avoid this ambiguity, even though distinguishing individual minerals can be more uncertain than measuring the total dust mass.

Figure

Figure

Additional ratios with respect to minerals other than quartz are shown in Figs. S3 to S6 of the Supplement.

Same as Fig.

The overestimated bulk fraction of combined phyllosilicates in the AMF
experiment at various locations within the Arabian Peninsula
(Fig.

Figure

All the experiments consistently underestimate the range of observed
mineral ratios (Fig.

The single particle measurements of mineral fractions at Tinfou cannot
distinguish the size distributions of the phyllosilicates and
feldspars, which are thus assumed to be identical

In a companion article

To evaluate the two experiments, we compiled measurements from nearly sixty studies that are distributed both near and far downwind of major dust source regions. In spite of this extensive compilation, many key sources remain undersampled. There are insufficient measurements to resolve the seasonal cycle of the mineral fractions and corroborate seasonal shifts of the dominant mineral calculated by the model that imply a change in source region. For example, kaolinite that is abundant in the Sahel dominates model deposition at Barbados during Northern Hemisphere winter, while an increase of emission in North Africa during the summer delivers more illite. In general, the uneven distribution of measurement sites and their limited duration imposes a large uncertainty that allows us to robustly evaluate only the most general features of the experiments.

Nonetheless, we show that the AMF method addresses key deficiencies of the
SMF experiment in comparison to measurements. In particular, AMF
phyllosilicates (that are nominally “clay” minerals) are most abundant at
silt sizes, while the silt fraction of quartz is reduced compared to the SMF
value and in better agreement with measurements. In spite of the unrealistic
behavior of the SMF method at silt sizes, both experiments show reasonable
agreement with measurements when the mineral fractions are summed over the
entire size range. This is because the emitted clay fraction in the SMF
experiment is large relative to the AMF experiment. This extra emission of
clay-sized phyllosilicates in the SMF simulation compensates for the absence
of these minerals at silt sizes. Similarly, the reduced fraction of emission
at silt sizes in the SMF experiment compensates for its excessive quartz
fraction. The fractional emission of clay and silt sizes in the SMF
experiment is based upon the local soil texture that is derived from
measurements of the fully dispersed, wet-sieved soil. However, the large
fraction of emitted clay-sized particles in the SMF method is inconsistent
with emission measurements that show a relatively small and regionally
invariant emitted clay fraction

The AMF method extends feldspar into the clay size range, consistent
with measurements. However, the bulk mineral fractions of carbonates,
gypsum and iron oxides are underestimated by both methods. The common
bias suggests an origin within the MMT fractions, although the aerosol
measurements themselves are infrequent and subject to uncertainty.
The underestimation of iron oxides may also result from the exclusion
of goethite from the MMT, a mineral that is a source of aerosol iron

Both the SMF and AMF experiments reveal a smaller range of mineral
ratios compared to the observations. This is partly a consequence of
model resolution that is insufficient to resolve strong spatial
contrasts in mineral fractions near isolated source regions. In
addition, spatial variations of soil mineral composition are reduced
by the MMT that consists of a single average value for all examples of
the same arid soil type. Common features of the AMF and SMF mineral
fractions at clay sizes are a useful test of the MMT, because the
emitted fractions in both experiments are unmodified by
reaggregation. Recent studies have proposed refinements to the MMT
based upon a greater number of soil measurements and inclusion of
additional minerals such as goethite, chlorite and vermiculite

Errors may also arise from our apportionment of the emitted silt to
the transported size bins. The AMF method currently apportions silt
emission using size-resolved measurements of individual minerals after
transport to Tinfou, Morocco. Evaluation of the model mineral
fractions suggests that prior deposition has preferentially removed
the largest particles

This study is a step toward calculating the influence of aerosol mineral
composition upon climate, including radiative forcing, physical and chemical
transformation during transport and aerosol solubility, among other
processes. While the global distributions of quartz and phyllosilicates like
illite and kaolinite are probably the best characterized by measurements,
other minerals with important climate impacts are subject to fewer
constraints. This is especially true for minerals like montmorillonite (a
member of the smectite group) and feldspar that are subject to fewer
measurements, resulting in an uncertain spatial distribution despite these
minerals' potential importance for ice nucleation

Despite the extensive compilation of measurements presented in
Table

We designed the evaluation of the SMF and AMF experiments to emphasize the differences between two methods of calculating aerosol mineral content. We compare mineral fractions rather than the absolute concentration of individual minerals to remove the effect of our uncertainty about the magnitude of global dust emission. Similarly, we relax the model winds toward reanalysis values so that the model mineral fractions are more strongly dependent upon the calculated fractions at emission rather than possible errors in aerosol transport.

Uncertainty of evaluation also results from sampling, including the
occasional departure of the measurement duration from the monthly averages
archived by the model. There are two general cases. In the first case, the
measurements represent an average over a duration of a month or longer and
can thus be compared directly with the archived model output. The measured
quantity in this case is typically deposition. For this example, we calculate
the standard deviation (SD) of the model, using the nine values available
from the 9 years simulated by each experiment. The SD allows us to estimate
a distribution of possible model values that can be compared to the single
measured value. That is, we are asking whether the measured value is
consistent with the model distribution. This allows a consistent treatment of
measurements that are both within and beyond the range of years corresponding
to our experiments. The model mean and SD of the mineral fractions are fitted
to a beta distribution that is commonly used to represent values that are
bounded between zero and one

In the second case, we have a measurement-like concentration whose duration
is less than the single month used to archive model output. In most examples,
we have multiple measurements from which we can estimate a time average and
standard error for comparison to the model. If these measurements are
confined to a single month, then we interpret the time average as an estimate
of the monthly average that can be compared to the model output. The
uncertainty of this average is estimated using the standard error

There are a few examples where daily measurements (or more
generally, measurements over sub-monthly durations) are scattered
over a much longer period. In some cases, the precise date of
measurement is unknown

A more rare case is where we have a measurement for only a single
day

There are a number of assumptions that go into our calculation of
measurement uncertainty. For example, Eq. (

We thank Paul Ginoux, Konrad Kandler, Jasper Kok, Natalie Mahowald,
Sergio Rodríguez, Rachel Scanza and Yves Balkanski for helpful
conversations. This article was improved by the thoughtful comments
of the reviewers. This research was supported by the National
Science Foundation (ATM-01-24258), the Department of Energy
(DE-SC0006713), the NASA Modeling, Analysis and Prediction Program
and the Ministry of Economy and Competitiveness of Spain through the
POLLINDUST (CGL2011-26259). NCEP Reanalysis winds were provided by
the Physical Sciences Division at the National Oceanic and
Atmospheric Administration Earth System Research Laboratory via