Airborne particles of mineral dust play a key role in Earth's climate system and affect human activities around the globe. The numerical weather modeling community has undertaken considerable efforts to accurately forecast these dust emissions. Here, for the first time in the literature, we thoroughly describe and document the Air Force Weather Agency (AFWA) dust emission scheme for the Georgia Institute of Technology–Goddard Global Ozone Chemistry Aerosol Radiation and Transport (GOCART) aerosol model within the Weather Research and Forecasting model with chemistry (WRF-Chem) and compare it to the other dust emission schemes available in WRF-Chem. The AFWA dust emission scheme addresses some shortcomings experienced by the earlier GOCART-WRF scheme. Improved model physics are designed to better handle emission of fine dust particles by representing saltation bombardment. WRF-Chem model performance with the AFWA scheme is evaluated against observations of dust emission in southwest Asia and compared to emissions predicted by the other schemes built into the WRF-Chem GOCART model. Results highlight the relative strengths of the available schemes, indicate the reasons for disagreement, and demonstrate the need for improved soil source data.
Airborne mineral dust particles play a key role in Earth's radiative budget,
weather and climate patterns, and biogeochemical processes
Over the past several decades, numerous dust emission and transport models
have been developed for forecasting and research purposes
First, we present a brief history of relevant model development. GOCART was
originally designed as a standalone, offline aerosol model driven by
assimilated meteorological fields
In 2011, researchers from the Air Force Weather Agency (AFWA), now designated
the 557th Weather Wing, and Atmospheric and Environmental Research, Inc.
(AER) began to investigate the WRF-Chem GOCART source code after noting
multiple unexpected simulation pattern results for dust emission in southwest
Asia. Closer inspection revealed issues with the dust emission scheme, which
rendered the original GOCART model dust output invalid under certain
environmental conditions. As a result, an alternative dust emission scheme
option was developed to augment the WRF-Chem GOCART code. Several journal
articles briefly discuss the use of the AER and AFWA modifications
To support the objectives of this paper, we provide a full documentation of
the GOCART-WRF dust emission scheme, including changes that have been made to
the code since
The paper is organized as follows. In Sect. 2, a brief background on the physics of dust emission is provided. In Sect. 3, the three dust emission schemes included in the WRF-Chem model are described. In Sects. 4 and 5, the model configuration and data analysis methods are described. In Sects. 6 and 7, the results of the study are presented and discussed. Conclusions are presented in Sect. 8.
Soil particles mobilize when lift, drag, and impact forces overcome the
gravitational and interparticle cohesive forces holding them to the soil bed
At present, there are three different dust emission schemes built into the
WRF-Chem model: the original GOCART-WRF scheme (
The version of the dust emission scheme originally described by
The original GOCART dust emission scheme is popular with the broader modeling
community because it does not require difficult-to-obtain soil or surface
characteristics to run (e.g., soil composition, micro- or macro-scale terrain
roughness, vegetation type and spacing, and soil aggregate strength).
Instead, geographic variability in substrate erodibility is fixed by a
simple, topographically based, internally calculated source function.
Erodible soil makeup is then fixed to a constant mix of sand, silt, and clay.
Wind speed, soil moisture, air density, and generalized soil traits are the
only necessary inputs for its dust emission flux calculation, and these are
determined from variables readily available in most numerical weather models.
This standalone nature of the original GOCART dust model has made it an
attractive choice for research and operational centers in need of regional-
or global-scale dust products
We first summarize the original GOCART dust emission scheme as it was
documented by
The threshold wind speed
Curiously, this means that the value of the correction factor varies from 0 to 1.2, equaling 1 at a soil moisture content of 10 %. This effectively treats the threshold wind speed for dry soil, calculated in Eq. (3), as if it were for soil having a moisture content of 10 % and could result in adjusted threshold wind speeds that are actually below the dry soil calculated wind speed for very low soil moisture conditions.
Dust mass flux values,
The GOCART-WRF dust emission scheme was first incorporated into WRF-Chem
version 3.2. and is called by setting There is a change in the number of dust emission size bins (now five) and the
size range for those bins (now 0.1–20 A precalculated source strength function A simplification of soil makeup is incorporated into the dust emission
flux (Eq. 1). All alluvium available for lofting is assumed to have a constant
distribution of 50 % sand, 25 % silt, and 25 % clay. The Another change is the addition and later removal of a tuning constant which multiplies
the emitted dust mass by 0.2 as it is being added to the first atmospheric
model layer. This tuning constant may produce unexpected results because it
does not alter the dust emission flux values output to the WRF-Chem history file,
even as it substantially reduces dust entrained into the atmosphere. The tuning
constant is present in versions 3.3–3.8 but is not present in versions 3.2, 3.2.1, and 3.8.1–4.0.1. The dimensional proportionality constant, Soil moisture values passed in by the WRF-Chem framework are converted
from volumetric water content ( The threshold soil moisture value used to restrict dust lofting in
Eq. (3) was set to 0.2 in WRF-Chem versions 3.2–3.4.1 but later changed to
0.5, bringing the value into agreement with The most substantive change in the GOCART-WRF dust emission scheme
relative to the description in where The switch to this revised scheme improved the ability of the GOCART model to
reproduce the known behavior of small diameter particles – specifically by
requiring higher threshold wind speeds for fine-particle mobilization. The
revision, therefore, produced empirically improved results. From a physical
standpoint, however, motivation for the use of the MB95 equation is strained
The change from
Modifying
The AFWA scheme is based on a modified version of the MB95 saltation-based
dust emission function. In the AFWA scheme, dust emission is handled as a
two-part process, wherein large particle saltation is triggered by wind shear
and leads to fine-particle emission by saltation bombardment. The equations
for the AFWA scheme are derived in terms of friction velocity,
Saltation particle size bins and their associated attributes.
Particle sizes are presented here in
Saltation processes for a given size bin initiate and cease during the
simulation as
Dust particle size bins and their associated attributes, presented
here in
In order to provide the gravimetric water content (
Once time-varying
Estimated contributions of each saltation size bin to total saltation flux
(
Bin-specific values of
Saltation bin-specific weighting factors are then found by taking the ratio
of
The total streamwise horizontal saltation flux is then computed
via
To estimate the bulk emission flux of dust (
Once total dust emission (
As with the GOCART-WRF scheme, the emitted dust particles are released into the lowest atmospheric model level for dispersion according to their respective size bins.
Four optional tuning parameters, three alternate input dataset channels, and
an optional modification to the
Schematic diagram of AFWA dust emission scheme and required inputs. The black diamond marker indicates that the parameter varies spatially and temporally. The black circle marker indicates that the parameter varies spatially, and the hollow diamond marker indicates the term is related to a particle size bin. See comprehensive variable list in Appendix B for variable definitions.
Optional tuning parameters and binary configuration flags (
An error in the number and distribution of saltation size bins was made
during the implementation of the AFWA scheme code into the WRF-Chem baseline.
Current and legacy versions of the AFWA scheme (WRF-Chem versions
3.4–4.0.1) assume nine saltation size bins (Table 1), including one clay-, five
silt-, and three sand-sized bins. Bins 7–9 are sand-sized bins with
effective diameters of 69, 131, and 250
WRF-Chem's third standard dust emission scheme, commonly referred to as the
University of Cologne (UoC) scheme, is activated by using
The UoC scheme follows the same general approach as the AFWA scheme. Both schemes simulate dust emission by first calculating a threshold friction velocity for particle saltation, then using that threshold friction velocity to determine saltation flux, and finally calculating emissions of dust particles caused by saltation processes (e.g., bombardment), capturing the general process of dust emission more fully than the GOCART-WRF scheme. Both schemes also use the same size-resolved dust emission bins to pass emitted dust fluxes to the WRF-Chem transport routines. The more sophisticated UoC sub-options also use size-resolved saltation particle bins to evaluate dust emission from saltating particles of different sizes.
The calculation of the threshold friction velocity for initiation of particle
saltation used by the UoC scheme is physically based and of significantly
different form, compared to the semi-empirical MB95 function used in the AFWA
scheme but has similar output in terms of calculated threshold friction
velocity (
After establishing the dry soil threshold friction velocity (
Vegetation fraction (
Once the corrected threshold friction velocity (
Note that, until the release of WRF-Chem version 4.0, there was an error in the implementation of this equation (discussed in Sect. 3.3.2).
The two differences in comparison with the AFWA scheme are (1) an adjustment
for vegetated fraction of the surface
In all UoC sub-options, just as in the AFWA scheme, the saltating particle
load in each size bin is also dependent on the fraction of the parent soil
consisting of particles in that size bin, and on the source strength
function. Source strength is again handled using the dust source strength
parameterization (stored as variable
Once the saltation fluxes are calculated, the next major step in the scheme
is calculating dust emission flux from the saltation flux,
(
S01 derives and uses the most complex form of the process, described as
Eq. (52) in S01. The parameterization includes effects of soil particle
aggregation, parent soil particle size distribution, saltating particle size
distribution, and soil plastic pressure, among other soil attributes.
S04 simplifies the scheme for estimating the dust emission from saltation
collisions by fixing several of the free variables in Eq. (25) which were not
readily available in measurements, including setting the collision angle to
15
S11 further simplifies the scheme by calculating dust emission based on a
single integrated saltation flux, rather than based on fluxes of saltating
particles in each individual saltation bin (and setting
This S11 approach is similar to the AFWA scheme, which integrates saltation flux across all saltation particle size bins (Eq. 13) and calculates a total dust emission from a total integrated saltation flux (Eq. 14). The two approaches differ in that the AFWA scheme sums the mass of all dust fluxes and then apportions the dust into size fractions based on a breaking function (Eq. 15). The simplified S11 sub-option, however, allows the dust particle size distribution to be based on parent soil type (Eq. 28).
In S01 and S04, the size-resolved dust emission is calculated by integrating
dust emissions of each dust bin over all saltation bins. During this step, an
additional factor of
This factor does not appear in the papers that document these schemes (S01, S04, S11) and may be in error; however, since the correction effectively reduces the surface area from which both sand particles and dust particles can be emitted, application of the correction twice (i.e., once for saltation and once for dust emission) may be physically valid.
The S11 sub-option yields size-resolved dust emission
In all UoC schemes, the total dust emission,
The effect of the more sophisticated approach in the UoC scheme is to make
both the saltating and emitted dust particle size distributions sensitive to
parent soil particle size distribution in S01 and S04 and to make the emitted
dust particle size distribution sensitive to parent soil particle size
distribution in S11. The approach makes the UoC scheme the most
physically based of the WRF-Chem dust emission schemes. Input data
limitations restrict the benefit of these sophisticated options, however.
Measurements of these soil characteristics are generally unavailable,
particularly over mesoscale domains (on the order of 10 km grid spacing), an
issue noted in the Shao publications. For example, the degree of soil
aggregation, used in the UoC scheme as the fully disturbed and
minimally disturbed soil particle size distribution, is not widely measured
or widely available in soil databases, nor is the soil plastic pressure.
Within WRF-Chem, the soil plastic pressure is simply set to a constant and
must be tuned to match local soil conditions. The particle size distributions
are derived based on a conversion between the soil particle size information
for the surface layer of soil (0–30 cm) originally derived from the FAO-SMW
soil dataset by
Dependence on the other key soil parameter, soil plastic pressure, controls
the mass ejected during bombardment collisions. In the Shao papers, test
cases are run to determine the best fit for the soil plastic strength based
on observational dust emission data, along with a dimensionless tuning
coefficient,
Similar to the GOCART-WRF scheme, we note that there are several
discrepancies between the code realization in WRF-Chem and the documentation
published in the literature in S01, S04, and S11. Again, we document these
here for the benefit of the community:
The equation used to calculate the saltation flux Changing the order of operations from how it is documented in S11, Given reasonable friction velocities, the effect could change the saltation
flux by a factor of 2 or more, resulting in substantial impacts on output. The equation used to calculate the threshold friction velocity for
particles in each saltation bin size, The implementation of the code appears to include the vegetation
coverage correction factor, The documentation for the earlier UoC models (S01 and S04) indicates
they use different equations for calculating saltation flux based on current
wind speed and threshold velocity than those used in S11. These equations are
of similar form and would produce similar saltation flux output to what would
be produced by the equation described in S11 (see S01, Eq. 23, which is derived
from We note that the number and character of the soil classes being composited
to determine the free dust fraction at particle sizes vary in the Shao
publications from 3 (S01, Table 2) to 12 (S04, Table 1) to 4 (S11, Table 2). As
implemented in the WRF-Chem model, the 12 soil texture classes of S04 are
applied to all three UoC sub-options. We also note a change in the number of dust size bins used to pass emitted
dust from the UoC scheme to the WRF-Chem transport routines between versions.
Four size bins with diameter ranges of
The original derivation of the UoC model handled saltation bombardment
and aggregate disintegration mechanisms separately (see derivation of Eq. 52 in S01,
Sect. 5), as opposed to handling all dust emission in a single bombardment-like process as is done in the AFWA scheme. The UoC scheme for calculating dust emission flux from saltation flux
(e.g., captured by Eq. 52 in S01, Eqs. 6, 7, and 11 in S04, and Eqs. 11 and 34 in S11)
depends on relatively sophisticated knowledge of the parent soil, including the
soil particle size distribution (the only term which the AFWA scheme also
depends on), measures of the degree of soil disturbance (e.g., captured in
The UoC scheme incorporates a correction factor in the calculation of
saltation flux for soil vegetation coverage. This factor has modest impacts on
results, and our test case indicates its utility may suffer from low-quality input data. The UoC scheme incorporates a second correction factor in the calculation
of threshold friction velocity for non-erodible roughness elements (i.e., a drag
partition correction), which is determined from the vegetation coverage layer.
We use WRF-Chem version 3.8.1
Domain map for the WRF-Chem simulations with color shading showing the waterbodies and elevation as indicated by the color bar. The region of dust emissions we focus on is just right of center in the Syrian Desert on both sides of the Iraq–Syria border.
Atmospheric dust was initialized using a “cold start” approach (i.e., the
dust concentration in the atmosphere was initialized as zero everywhere). The
model background chemistry for other aerosol species was generated using the
GOCART simple option in WRF-Chem. Background sea salt emissions were based on
the lowest model level wind speeds over the oceans (
WRF-Chem physics and chemistry parameterizations.
Three schemes for deriving dust emissions in WRF-Chem (GOCART-WRF, AFWA, and UoC – discussed separately in Sect. 3) are tested, and we compare the results below. All three dust emission schemes tested were run in the “default” configuration supplied with WRF-Chem version 3.8.1 release to permit the most straightforward comparison, with all constants set as supplied in the code release and described above in documentation for each scheme. For the purposes of this paper, we chose to make comparisons to the moderately simplified version of the UoC scheme described in S04.
For intercomparison of model results with remote sensing data, simulated
atmospheric extinction coefficients are calculated for the 550 nm wavelength
using the WRF-Chem optics routines
Integrated column AOD is sampled from the model for comparison with satellite remote sensing observations collected from the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP) instrument at the grid point nearest to observational geographic coordinates (lat/long). For comparison with CALIOP data, coordinates used represent the midpoint of the 15 along-track samples that are averaged to produce a single AOD estimate. Since samples are collected every 333 m by CALIOP, actual observations extend 2.5 km from the midpoint in each direction along track.
The test event selected for our emission scheme intercomparison was a dust storm in southwest Asia forced by a large-scale synoptic event. We chose this location because we expect that the conditions the AFWA scheme was created for frequently prevail there. Specifically, spurious dust lofting under light wind conditions has been noted in this region in WRF-Chem runs with the GOCART-WRF dust emission scheme activated, as discussed in Sect. 3.1. The atmospheric dust plumes observed by MODIS AOD satellite remote sensing during this event appeared to originate in western Iraq and Syria, qualitatively indicating a large, possibly dominant, role for dust emission from this region during the event.
While we compare remote sensing and simulation results throughout the event, we focus most of our analysis on the time period between 06:00 and 23:00 UTC on 25 January when a classic wintertime shamal moved across the analysis domain, causing emission and lofting of dust from the Syrian Desert. During a shamal, a cold front sweeps across the Arabian Peninsula allowing a high pressure to build in from the northwest and strengthen across Saudi Arabia. The synoptic pattern forces strong northwesterly surface winds to blow across the Syrian Desert and often lofts large quantities of dust into the atmosphere.
We characterize the synoptic evolution and evaluate the meteorology of the
WRF-Chem simulation using the Climate Forecast System Reanalysis
The synoptic environment during the time period surrounding the dust
emission event. Blue lines represent 700 hPa geopotential height, shading
represents 850 hPa temperature, and vectors represent 925 hPa winds. Panels
To evaluate the realism of the modeled synoptic evolution, we compared the
variables used to characterize the synoptic environment from WRF-Chem
(Fig. 3a–d) with the independent CFSR data (Fig. 3e–h). The synoptic evolution
produced by the WRF-Chem model was very similar to the one in the CFSR,
indicating that WRF-Chem performed adequately in simulating the meteorology.
Further comparisons to radiosonde data (not shown) indicated WRF-Chem was
able to adequately reproduce the observed atmospheric wind and temperature
profiles
We utilize 1 km resolution true-color and dust-enhanced satellite imagery
derived from MODIS data to qualitatively assess the general origin and extent
of the dust plumes. Image dust enhancement was performed using a processing
algorithm by
We use version 4 (V4) of the level 2 (L2) vertical feature mask data product
(CAL_LID_L2_VFM-Standard-V4-1) from CALIOP on board the Cloud-Aerosol Lidar
and Infrared Pathfinder Satellite Observation (CALIPSO) mission to identify
atmospheric aerosol observed in the modeled domain
Observations of the lofted dust plume collected at 10:00 UTC on 25 January 2010
by the MODIS sensor including the
Results from the three simulations (Figs. 5–7) demonstrate substantial
differences in outcomes between the GOCART-WRF scheme (
Shaded maps of modeled AOD for the GOCART-WRF (column a), AFWA (column b), and UoC (column c) schemes. Timestamps are indicated at the top of each image and are the same across the rows. Transects of AOD are also placed as overlays on each plot, with the three adjacent transects representing observed AOD from the CALIOP data (left transect line), locations along the transect where CALIOP observations are heavily impacted by cloud cover and retrieval does not represent full column AOD (center transect line), and modeled full-column AOD along the transect (right transect line).
The collection of these simulations clearly demonstrates that the GOCART-WRF scheme produces the largest atmospheric dust content, and that the dust lofts from across the widest area, including intense emissions from the Syrian Desert in eastern Syria, Jordan, and western Iraq and lower intensity emissions in the northern Arabian Desert areas of southern and western Iraq and northern Saudi Arabia (Fig. 7). The dust emissions occur over a wider area and continue temporally longer than they do in the other schemes, including in areas experiencing lower wind speeds. This outcome is consistent with the spurious dust lofting noted by earlier works. The result of these large-scale emissions is substantial AOD over large areas of the model domain (Fig. 5). The excessive area experiencing dust lofting is largely expected given the treatment of the threshold wind speed discussed in Sect. 3.1.
Aerosol extinction profiles at 550 nm along the CALIOP transects for each of the six overpasses shown as integrated AOD in Fig. 5. Row 1 represents observed CALIOP data, with light gray indicating clouds, and dark gray indicating areas beneath clouds where no data are available. The remaining three rows represent modeled data, with GOCART-WRF in row 2, AFWA in row 3, and UoC in row 4.
Modeled dust emissions from the GOCART-WRF
The AFWA and UoC schemes both produce much more localized emissions and emit dust only under the higher wind conditions present early on 25 January (Fig. 7). Emissions in the AFWA scheme originate from the Syrian Desert in southern and eastern Syria, western Iraq, and eastern Jordan but are limited beyond this domain and of much lower intensity than seen in the GOCART-WRF scheme. These result in AOD patterns consistent with a “pulse” of dust emission as the front passes over the Syrian Desert. The pulse is then advected eastward and northward out of the model domain (Fig. 5). The spatial configuration of emissions is still more localized for the UoC scheme, restricted to intense emission sites in the Syrian Desert, primarily in southern Syria, but also in extreme eastern Jordan and extreme western Iraq. The modeled AOD resulting from the highly localized emission of the UoC scheme is then an intense pulse with relatively hard boundaries. Similar to the AFWA scheme, this is advected east and northward out of the domain but covers a much smaller spatial extent during this time.
Compared to the spatial extent of the dust plume seen in the dust-enhanced MODIS observations (Fig. 4), the modeled AOD in the AFWA scheme (Fig. 5) produces the best match to the AOD seen in the cloud-free region within the MODIS observations, in this particular test case. Modeled AOD shows too small a spatial extent in the UoC scheme and too large a spatial extent in the GOCART-WRF scheme (Fig. 5). This single test case comparison does not imply that any of the three dust emission schemes is superior in all cases. This result, however, provides the basis for investigating the reasons for the particular model behavior in the discussion that follows.
More detailed comparisons of simulated and observed dust in the atmosphere are presented using the CALIOP lidar data. Total column AOD is presented along the CALIOP tracks in Fig. 5. The parallel transects represent the observed (left) and simulated AOD (right) with cloud cover that restricts CALIOP retrieval of full-column AOD indicated in the center transect. Note that observed and simulated AOD should only be compared in areas not impacted by cloud cover. Unfortunately, high observed AOD frequently occurs in close proximity to cloud cover, and none of the available CALIOP transects directly sample the main dust plume of this event near the time of peak emissions. While these limitations hinder a robust comparison, a general result is that the GOCART-WRF scheme tends to produce higher AOD along the CALIOP transect than observations show (e.g., Fig. 5, row 3), while the AFWA and UoC schemes both show more limited AOD along the transects which appear smaller in extent than suggested by observations. All schemes appear to under predict the highest values of observed AOD. Closer examination of this in profile format is needed to better assess agreement.
Modeled and observed aerosol extinction profiles are presented in Fig. 6. A combined plot representing several CALIOP observations is presented in the first row. The plot is based on vertical feature mask data (to show clouds) and extinction profiles, where available. Optically thick clouds are masked in light gray and the area underneath optically thick clouds (no data) is masked in dark gray. This more clearly shows the substantial limitations on available data in the lower atmosphere imposed by cloud cover and the reason for limited observations of high total column AOD in the transects shown in Fig. 5. The extinction coefficients presented, both in these observed data and in the model profiles below, may be reasonably thought of as being caused entirely by dust because aerosol extinction is overwhelmingly attributed to mineral dust in both CALIOP aerosol layer product and in modeled data.
The modeled extinction profiles presented in rows 2–4 indicate that the
location of dust in the atmosphere is largely consistent between the three
dust emission scheme configurations but that the amount of dust in the
atmosphere differs substantially, with the most dust produced by the
GOCART-WRF scheme and the least dust by the UoC scheme. The altitude and
spatial placement of the modeled atmospheric dust (as indicated by extinction
coefficients) along CALIOP passes collected at 11:00 UTC on 24 January 2010, 23:00 UTC on 24 January 2010,
00:00 UTC on 26 January 2010, and 01:00 UTC on 26 January 2010
all appear consistent with observations, though the observed atmospheric
extinction is higher than the amount present in all simulations. In these,
the overall dust entrained into the atmosphere in the GOCART-WRF scheme, even
though it is emitted from far too large a spatial area, is the best match for
observed extinction profiles, in terms of magnitude. Limited observations due
to cloud cover make the 10:00 UTC on 25 January 2010 pass challenging to assess.
Modeled dust at 23:00 UTC on 26 January 2010 is consistent with the other four
time steps, in that altitude and spatial placement of the model dust
(extinction coefficients) along the southern end of the transect broadly
matches observations but differs in that the GOCART-WRF and AFWA schemes
exhibit much stronger extinction profiles in the central part of the transect
from 32.5 to 27.5
The
Figure 8 compares simulated 8 h average 550 nm AOD centered at 10:00 UTC on 25 January 2010 to the MCD19A2 MODIS AOD product from 25 January 2010. The effect of clouds on the MODIS AOD retrieval is evident, as much of the AOD in the image is masked out. A regional peak in AOD is observed near the border of Iraq and Saudi Arabia. The general patterns of average AOD simulated for the same time period by the GOCART-WRF scheme are broadly consistent with the MODIS AOD product in the southern part of Iraq and over the Persian Gulf. An area of high AOD in northern Iraq is challenging to compare to observations due to a lack of data in much of that region. Simulated AFWA scheme AOD is too strong over eastern Iraq and also appears to be placed west of the observed plume, perhaps due to a mismatch in timing of emission and therefore less downwind transport, but still captures the extent of the plume across the southern half of Iraq towards Kuwait. Again, high AOD in northern Iraq is difficult to assess. There is a mismatch between the high AOD modeled by the AFWA scheme in northwestern Iraq and observations, but a lack of data just east of the simulated plume location prohibits assessing whether there is simply a small temporal mismatch. There is less agreement with the UoC scheme, which produces several localized, high AOD values over Syria, Jordan, and western Iraq instead of the broader AOD patterns generated by the other two schemes.
We primarily intend our test case data to be a tool for discussing
differences between the three WRF-Chem dust emission schemes. We therefore
explore the reasons for the differences between these emission schemes in
greater detail and plot several static and intermediate model variables as
diagnostics to illuminate the various sources of the large differences in the
spatial extent and intensity of the modeled dust emissions and to identify
highly sensitive model parameters. Relevant terrain attributes, including
Relevant test domain terrain attributes, including source strength
(
Simulated wind speed
Values of intermediate variables used in the calculation of dust
emissions by the three different emission schemes, with the GOCART-WRF
scheme in the top row, the AFWA scheme in the middle row, and the UoC scheme
in the bottom row. All images reflect model state at 11:00 UTC on 25 January 2010.
The theoretical dry soil friction velocity threshold for saltation of
grains having diameter 60
The UoC vegetation correction function, squared to account for the application of the multiplier in both the saltation and emission flux calculations. Areas of dark gray are water bodies, and areas void of color are areas masked out for vegetation in the source strength function.
Here, we are particularly interested in explaining the reasons for the
differences in spatial coverage of dust emission in the UoC scheme, relative
to the AFWA scheme. Reasons for spurious dust lofting at low wind speeds in
the GOCART-WRF scheme are well documented in Sect. 3.1 and by earlier
papers
We begin our analysis by calculating dry soil threshold friction velocity required for initiating particle mobilization for each of the three dust emission schemes. The dry soil threshold parameter for these schemes only varies as a function of particle size (i.e., it does not vary spatially); however, we provide results in mapped display (Fig. 11, column 1) for ease of discussion with respect to the soil moisture and roughness correction factors. Resultant dry soil thresholds for given particle sizes are shaded everywhere the dust source function is non-zero.
Direct comparison between the GOCART-WRF scheme and the other two schemes is
not possible since the GOCART-WRF scheme only considers dust-sized particles,
but for completeness we determine the dry soil threshold velocity for a grain
diameter of 16
All three dust emission schemes include a correction for the threshold
friction velocity parameter based on the soil moisture. This correction
factor is shown in Fig. 11, column 2. The general equation for calculating
this correction in the AFWA and UoC schemes is identical
The soil moisture correction in the GOCART-WRF scheme is quite different, and
its value varies from 0 to 1.2, with values near zero for soils of very low
moisture content. The values
In the UoC scheme, the moisture-corrected threshold friction velocity is
further modified by a roughness correction (Eq. 18), calculated based on
vegetation coverage (Eq. 19). Vegetation fraction,
Threshold friction velocities with all corrections applied are then shown in Fig. 11, column 5. These fields, which can be compared against the values from column 3 that have only the moisture correction applied, clearly show that the roughness correction increases the threshold friction velocity across the western Iraq area in the UoC scheme, while leaving the threshold friction velocity similar to the AFWA scheme in southern Syria.
Next, saltation flux for the denoted saltation particle size is calculated
from the WRF-Chem simulated wind speed or friction velocity and the threshold
friction velocity. This is shown in Fig. 11, column 6, for particles of
60
Values from Fig. 11, column 6 (calculated for all relevant particle sizes
associated with a given scheme), are converted to predicted emission fluxes by
considering the availability of erodible substrate, which is captured in all
schemes in some form by the topographically derived source function (
The
The final dust fluxes presented in Fig. 7, row 2, incorporate additional
factors. The GOCART-WRF and AFWA schemes amount to simple multiplications of
the source terms and theoretical fluxes, with different methods for handling
the parent soil particle size distribution and a small additional correction
factor (
We conclude from this analysis that the primary cause of the differences in
dust emissions between the AFWA and UoC schemes is the combined effect of
multiple terms. Emissions in western Iraq for the UoC scheme are primarily
restricted by the roughness correction applied to the threshold friction
velocity (Eqs. 18 and 19) with influence from the saltation flux coding
error and the vegetation correction on the overall emission magnitude. These
roughness and vegetation effects ultimately trace back to the vegetation
fraction,
The finding that the vegetation layer is essentially controlling the spatial
extent of dust emissions in the UoC scheme highlights an important fact –
dust emission models are highly sensitive to terrain condition data inputs,
which are determined from notoriously sparse datasets and (as discussed by
Aside from improving vegetation coverage or soil composition data, we note that several parameters could be tuned to attempt to better match behavior between the schemes or better match model behavior to observations. The UoC scheme is particularly sensitive to the soil plastic pressure, and this variable is set to a constant for the entire model domain. Tuning this variable can result in matching the dust emissions of select regions but not across the entire model domain, suggesting this parameter should be dependent on soil type and set using a spatially varying input dataset.
The AFWA dust emission scheme for WRF-Chem is fully documented in the literature for the first time here. This emission scheme represents a substantial advance in the physical realism of dust emission modeling over the GOCART-WRF emission scheme. Key improvements to model algorithms permit saltation flux, caused by aerodynamic entrainment, to be modeled separately from dust emission, largely caused by bombardment and disaggregation processes. Output from the model in a test case is shown to broadly match the spatial distribution and intensity of dust emissions during a wintertime shamal event in southwest Asia.
Analysis of the code and documentation available for the other dust emission schemes highlights several discrepancies between documentation and code implementation, as well as several changes in code implementation across WRF-Chem versions that had not previously been documented. In particular, a recently corrected error in the implementation of the UoC scheme (see Sect. 3.3.2) may have resulted in emissions from the implementation present in WRF-Chem prior to version 4.0 that were approximately an order of magnitude lower than would be expected from the parameterization that should have been included.
Comparing the parameterization approach of the AFWA scheme to the UoC scheme, as implemented in WRF-Chem version 3.8.1, highlights that the two models are similar in many ways. Though the processes included in the UoC dust emission scheme are potentially more physically complete, the AFWA model may have an advantage in mesoscale development due to its lower sensitivity to sparse and challenging-to-obtain soil and vegetation data. The most important future opportunities for improving both AFWA and UoC schemes appear to be related to the fixed input data on terrain properties. First and foremost, both schemes would benefit greatly from replacing the soil particle size distribution dataset and erodibility function with better observational data. UoC would also benefit from improved soil and vegetation coverage data and from a function to make soil plastic pressure tied to soil type or particle size distribution. A focus on collecting and synthesizing such wide-ranging data on Earth surface characteristics, however, will require a substantial, coordinated community effort.
The code used in this study (WRF-Chem version 3.8.1)
is included in the chemistry package of the WRF model, currently available
through
The results and discussion presented in our study explore use of the three currently available WRF-Chem dust emission schemes as they are presented in version 3.8.1; however, as highlighted in the text, there are some relatively easy-to-correct errors in the AFWA and UoC code that are worth examining further. Here, we assess the effects of the UoC saltation function order of operations error described in Sect. 3.3.2 (i.e., Eqs. 34 and 35) and use of an alternate configuration for the AFWA scheme saltation bins by rerunning our simulation with bug fixes applied for comparison.
For the UoC scheme, we correct the order of operations error in the UoC
saltation flux calculation (i.e., Eqs. 34 and 35). While this error was
corrected in WRF-Chem version 4.0 (released June 2018), the bug remains in
all previously released versions of WRF-Chem, including version 3.8.1. For
the AFWA scheme, we reran our simulation using an alternate saltation bin
configuration described in Table (A1) that better aligns with the mass
distributions recommended by
Difference in simulated 8 h mean AOD (centered at 10:00 UTC on 25 January 2010)
produced by the modified and original versions of
Simulated 8 h mean AODs (centered on 25 January 2010 at 10:00 UTC) from the original and altered UoC and AFWA version 3.8.1 codes were used to illustrate the effects of these changes. Figure A1 shows the calculated difference in 8 h mean AOD between the corrected and uncorrected versions of each scheme. The UoC scheme correction has little effect on the spatial extent of the dust plume but essentially doubles the AOD magnitude in regions where dust is present. Similarly, the use of the alternate saltation bins in the AFWA scheme has a relatively negligible effect on the location and extent of the simulated dust plume. However, in contrast to the UoC correction, the AFWA AOD differences are smaller and of mixed sign.
Based on these results, we recommend that model users consider the impact of the UoC saltation flux error when assessing published results from studies performed using the UoC scheme prior to the release of WRF-Chem version 4.0. The effects of the alternate saltation bin configuration on overall AFWA scheme performance are less clear. Optimal settings for the saltation arrays may be region dependent. Further analyses beyond the scope of this paper are still needed.
Alternate saltation particle size bin configuration and associated
attributes recommended for use with the AFWA scheme. Values are presented
here in
Variable list.
Variable list continued.
SLL and GAC developed the AFWA dust emission scheme. JDC supervised project execution of the AFWA scheme code development. GAC and SEP implemented the AFWA scheme code into the WRF-Chem framework. TWL conducted and post-processed the WRF-Chem case study simulations. SLL, CP, and TWL analyzed data and primarily wrote the manuscript. All co-authors critically reviewed the manuscript.
The authors declare that they have no conflict of interest.
Funding support for this project was provided by the U.S. Army Terrestrial
Environmental Modeling & Intelligence System Science Technology
Objective–Research (ARTEMIS STO-R) 053HJ0/FAN U4357509, “Dynamic Undisturbed
Soils Testbed to Characterize Local Origins and Uncertainties of Dust
(DUST-CLOUD)”, applied science research program sponsored by the Assistant
Secretary of the Army for Acquisition, Logistics, and Technology (ASA-ALT).
Access to CALIOP and MODIS data was provided by the NASA Earth Data Portal at