Reducing Resolution Dependency of Dust Emission Modeling Using Albedo‐Based Wind Friction

Numerical simulations of dust emission processes are essential for dust cycle modeling and dust‐atmosphere interactions. Models have coarse spatial resolutions which, without tackling sub‐grid scale heterogeneity, bias finely resolved dust emission. Soil surface wind friction velocity (us*) drives dust emission non‐linearly with increasing model resolution, due mainly to thresholds of sediment entrainment. Albedo is area‐integrated, scales linearly with resolution, is related to us* and hence represents its sub‐grid scale heterogeneity. Calibrated albedo‐based global dust emission estimates decreased by only 2 Tg y−1 (10.5%) upscaled from 0.5 to 111 km, largely independent of resolution. Without adjusting wind fields, this scaling uncertainty is within recent estimates of global dust emission model uncertainty (±14.9 Tg y−1). This intrinsic scaling capability of the albedo‐based approach offers considerable potential to reduce resolution dependency of dust cycle modeling and improve the representation of local dust emission in Earth system models and operational air quality forecasting.


Introduction
Atmospheric mineral dust impacts Earth's systems, human health, and global economies (Tegen and Schepanski, 2018;Li et al., 2018;Pi et al., 2020).Global dust models were developed more than two decades ago (Joussaume, 1990;Marticorena and Bergametti, 1995;Shao et al., 1996) to resolve spatial patterns and trends of aeolian processes in the dust cycle (Shao et al., 2011).These models are essential for palaeo-environmental reconstructions (Mahowald et al., 2010), dust-climate interactions (Albani et al., 2014;Shao et al., 2013) air quality applications, and dust-atmosphere interactions (Wilcox et al., 2010).These global dust cycle models have coarse horizontal resolutions.Applying a single mean wind speed to these large pixels can bias dust emission (F) model simulations depending on the sub-grid scale heterogeneity of the wind friction velocity (Raupach and Lu, 2004;Ridley et al., 2013).The inability of the coarse resolution to resolve the fine resolution processes, challenges the accurate estimation of F trends and patterns (Meng et al., 2021).Dust emission is dependent on horizontal sediment flux (Q) and the sandblasting process which is influenced by the entrainment threshold (u *ts ).The magnitude and frequency of processes that cause u *ts to be exceeded depend on the soil surface wind friction velocity (u s* ) transferred from above the largest roughness scale, typically controlled by vegetation.The physics of wind-blown F (Shao et al., 2011) are typically at the grain-scale (Raupach and Lu, 2004).These are point "support", following the geostatistical literature, where "support" is the volume of measurement (Kyriakidis and Yoo, 2005).For clarity, the u *ts is point support which should be compared with the point support u s* for compatibility, to calculate Q and F. Regional and global F models are driven by coarse resolution wind fields modeled over an area or pixel support.Despite being provided on a regular grid, these wind fields are not point support and are therefore incompatible with grain-scale physics and parameterizations (Gotway and Young, 2002) used in Q and F models (Leung et al., 2023;Meng et al., 2021;Raupach and Lu, 2004;Ridley et al., 2013).For clarity, we use model resolution to refer to different sized areal data or pixel support and in our work we do not use the terms "grid resolution" and "grid spacing" to avoid giving the impression that these data are point support.To address the problem of missing F due to the smoothing of the sub-grid wind maxima, a common approach is to employ a gridded Weibull distribution (Cakmur et al., 2004;Cowie et al., 2015;Grini et al., 2005;Menut, 2018;Tai et al., 2021;Zhang et al., 2016).Meng et al. (2021) developed and applied a method to harmonize F across simulations of different resolutions by generating offline F from fine resolution meteorological fields.Leung et al. (2023) proposed an alternative approach to derive a simple spatial map and upscale the spatial variability of F from fine to coarse resolution.
Aerodynamic roughness depends on wind speed and responds to roughness element height, width and configuration.As area increases, roughness elements are not linearly additive because of their mutual sheltering caused by their configuration.Consequently, aerodynamic roughness does not scale linearly with increasing area.Raupach and Lu (2004) showed incisively that the scaling problem of F models can be resolved by parameterizing hard-to-measure variables for example, wind friction velocity, with more widely available, area-integrated data for example, albedo which scales linearly with increasing area.This approach was implemented recently for the calculation of the areal soil surface wind friction velocity (u s* ) using a calibration with albedo which scales linearly over increasing area (Chappell and Webb, 2016;Ziegler et al., 2020;Zhou et al., 2024).The calibration is applied after the albedo is scaled to the desired resolution to overcome the non-linearity of upscaling u s* .This albedo-based approach was developed for use across scales from ground-based to Earth System Models (ESMs), wherever albedo data are available.It enables correctly upscaled, area-weighted estimates of u s* for calculating Q and F demonstrated using MODIS albedo daily, every 500 m across Earth (Chappell and Webb, 2016;Chappell et al., 2019;Hennen et al., 2022).This albedo-based approach improves regional dust model performance with better representation of roughness (LeGrand et al., 2023).This approach offers the opportunity to investigate the extent to which existing regional and global Q and F models are resolution dependent, constrained by the incompatibility of the grain-scale model and the areal wind fields.
This scaling issue at least partially explains why there are several orders of magnitude difference in updated global F estimates (29-4,313 Tg y 1 up to 20 μm particle diameter (Chappell, Webb, Hennen, Schepanski, et al., 2023;Huneeus et al., 2011)).Ridley et al. (2013) suggested that the fidelity of F modeling may benefit from better representation of these roughness and texture properties and consideration of their sub-grid distribution.The arising hypothesis is therefore that compatibility between areal wind fields and areal u s* should improve the performance of F modeling across scales (Chappell, Webb, Hennen, Schepanski, et al., 2023).

Theoretical Basis for Scaling Using Linearly Additive Surrogate Space-Time Data
Large uncertainty occurs in coarsely resolved modeling due to sub-grid scale heterogeneity (de Vrese and Hagemann, 2016).That heterogeneity is always much larger than any coarsely resolved model (Raupach and Lu, 2004).Consequently, processes occurring at fine resolution are typically described by relations and variables applicable at single points.Raupach and Lu (2004) denote these grain-scale models by f = f(v), where f is a flux for example, F, and v a vector of driving variables determining the flux for example, u s* .However, models require parameterizations between fluxes and variables averaged spatially over model unit areas (pixels) and temporally over model time steps that is, f = f (v) , where overbars denote space-time averaging.The scaling problem of how to obtain f (v) from f(v), is not solved by simply averaging, resampling or interpolating from one grid spacing to another (Leung et al., 2023;Ridley et al., 2013).Scaling requires the conversion of point support to area support (de Vrese and Hagemann, 2016;Kyriakidis and Yoo, 2005;Raupach and Finnigan, 1995;Van Looy et al., 2017).Raupach and Lu (2004) succinctly describe the scaling problem as finding where p(v) is the probability density function of v in a pixel.Following Raupach and Lu (2004), there are two cases of the scaling problem.The first case is relatively trivial where the grain-scale model f(v) is a linearly additive form of fluxes and driving variables.The second more challenging case is when the grain-scale model f (v) is non-linear, common in aeolian transport modeling.In this latter non-linear case, either the grain-scale model f(v) and averaged models f (v) have different functional forms, or the averaged quantities f ,v) are not related to the grain-scale by linear averages.The origin of the problem is the combined influence of non-linearity in f(v) and statistical variability in v (Equation 1).This causes the scaling problem to be dependent on the sub-grid scale heterogeneity through the p(v) of the driving variables v. Raupach and Lu (2004) described the most effective way to acquire sufficient spatially distributed information of sub-grid scale heterogeneity in p(v).Their recommendation was to constrain spatial patterns in parameters by using readily available data which is spatially correlated with hard-to-measure properties for example, u s* .Consistent with Equation 1, this approach amounts to estimating p(v) by using surrogates, for which space-time information is available to determine spatial patterns in v.We follow this approach using the albedo-based relation with u s* for scaling F modeling.

Dust Emission Modeling Using Areal Albedo-Based u s*
The flux f(v) of point support dust emission F (kg m 2 s 1 ) is reliant on grain-scale sediment flux Q (kg m 1 s 1 (Namikas and Sherman, 1997)) and calculated using the albedo-based approach (AEM) That flux has driving variables (v; Equation 1) over typically large (e.g., >11 km) areas with sub-grid scale heterogeneity in u s* , with the momentum remaining at the soil surface after removing the influence of roughness at all larger scales.The areal coupled parameter u s* /U h is retrieved at any scale by using the linear scaling relation with albedo (ω; Equation S1 in Supporting Information S1) with an estimation uncertainty of 0.0027 m s 1 (Chappell & Webb, 2016; Text S1 in Supporting Information S1).The Q AEM requires u s* enabled by multiplying u s* /U h with near surface areal mean wind speed (U) for a given height h (where U h is typically resolution dependent).The u *ts (d) is a grain-scale entrainment threshold wind friction velocity (m s 1 ).The u *ts (Marticorena and Bergametti, 1995) and its adjustment using H(w), a function of soil moisture (w) which inhibits entrainment (Fécan et al., 1999), are both described in standard workflows (Darmenova et al., 2009) and in Text S2 in Supporting Information S1.For simplicity, ρ a air density (1.23 kg m 3 ), g gravitational acceleration (9.81 m s 2 ), and c = 1 is a dimensionless fitting parameter.The classical Q TEM uses u * described in standard workflows (Darmenova et al., 2009) and in Supplement S2.The classical Q TEM approach requires the drag partition R (z 0 ,z 0s ) = u s* /u * which is poorly constrained with aerodynamic roughness lengths (z 0 ,z 0s ) typically (Zender et al., 2003) fixed over space and static over time (z 0 = 100 μm and z 0s = 33.3μm), which fixes R ≈ 0.91 and the magnitude of Q TEM is also over-estimated by using u * cubed (Chappell, Webb, Hennen, Zender, et al., 2023;Webb et al., 2020).

Dust Emission Modeling
The albedo-based dust emission flux (F AEM for particles <10 μm; kg m 2 s 1 ) is calculated as a function of soil clay content (Marticorena and Bergametti, 1995) We restricted clay% to a maximum value of 20% consistent with previous work (Marticorena and Bergametti, 1995).The Q AEM producing dust, is adjusted by the emitted dust fraction M for a given particle size fraction with diameter d which we calculated as 0.1 < d < 10 μm (Zender et al., 2003) by using M = 0.87.The F of a pixel is masked out if the soil surface is bare but frozen, which inhibits F(A f ) or in the presence of any snow coverage (A s ).Unlike existing F models, vegetation indices and fixed vegetation coefficients in the erodibility parameter E are unnecessary and not used (Text S3 in Supporting Information S1; Chappell, Webb, Hennen, Zender, et al., 2023).Furthermore, we do not apply preferential dust source masks to pre-condition the magnitude and geographical distribution of F. The classical dust emission modeling (F TEM ) used here is described in standard workflows (Darmenova et al., 2009) and included in Text S3 in Supporting Information S1.
In the absence of information to parameterize u *ts heterogeneity at the sub-grid scale, it has long been assumed fixed over space within surface texture type, and considered static over time.The ground is assumed to supply

Data for Implementing Dust Emission Modeling
The albedo-based approach provides areal estimated u s* /U h every day, every 0.5 km, across Earth's land surface.Albedo (MCD43A1 Collection 6) is available from MODIS aboard polar-orbiting satellites which cause incomplete coverage.However, the variation in albedo-based roughness at the daily-weekly scale is sufficiently small that we were able to smooth the available data weekly, to improve the coverage.MODIS albedo is areaintegrated and scales linearly (Case 1; Raupach & Lu, 2004).For a given day with 0.5 km pixels, albedo is reprojected to pixels at >0.5 km and then calibrated to provide areal estimates of u s* /U h (Equation S1 in Supporting Information S1) for F cal .The reprojection forces each albedo image with a given pixel resolution (e.g., 0.5 km) to be computed to a different (e.g., 5 km) pixel resolution which is equivalent to resampling using a bilinear interpolation.We repeat this process for all albedo images, for all resolutions in our study (0.5 km, 5 km, 11 km, 28 km, 55 km, 111 km, and 278 km) to examine F resolution dependency.For consistency with comparisons across these resolutions, we use a spatial restriction set to the largest 278 km resolution, applied to all smaller resolutions.
For consistency with the daily MODIS albedo data, we converted ERA5-Land hourly to daily data and used wind maxima to better approximate the gusts which cause sediment transport and dust emission.The coverage of snow ranges between 0 and 1 and we applied the MODIS Normalized Difference Snow Index (Hall et al., 2016) (MOD10A1 Terra, daily at 500 m).We used soil surface temperature with a threshold of 273.15K above which fluxes may occur.Soil clay content (250 m pixels) was represented with a digital soil texture map (Dai et al., 2019;Hengl et al., 2017).Consistent with the treatment of the albedo data, we reprojected these driving variables in the same way.Notably, for grid resolutions which were finer than the native grid resolutions (0.5 and 5 km) of the driving variables, we did not use any downscaling.We recognize that these driving variables are unlikely to scale linearly and therefore will not provide the most accurate values at different resolutions.Nevertheless, these reprojected driving variables were applied consistently to both F cal and F TEM .This approach enabled us to demonstrate the performance of scaling u s* and its relative impact on dust emission across different resolutions.

Global Seasonal Differences in Dust Emission With Resolution
The pattern and magnitude in the daily values, aggregated to mean monthly u s* normalized by wind speed (u s* /U h ) during 2022, is similar for each resolution up to 111 km (Figure 1a).The similar pattern indicates that on average aerodynamic roughness is largest (u s* /U h is smallest) during March-May than in any other season.In contrast, u s* /U h (for equivalence) in the classical approach is "tuned" to a large value by assuming a smooth land surface devoid of vegetation fixed over space and static over time (z 0 = 100 μm and z 0s = 33.3μm) which fixes R (z 0 ) ≈ 0.91 and therefore fixes u s* /U h (Figure 1b).
The linear upscaling of albedo calibrated to u s* /U h is evidently largely independent of resolution up to 111 km.Consequently, the Q AEM (Equation 2) is also largest during boreal spring (Figure 1c).More importantly, up to and including 55 km resolution, the Q AEM displays little resolution dependence all year.At 111 km and 278 km resolution, resolution dependency occurs in all months and the magnitude of that dependency is proportional to the magnitude of Q AEM (Figure 1c).These differences are also evident in the u s* /U h indicating that at resolutions of 111 km and above, the information content in the roughness has changed.This is very likely caused by differences in the sources of landscape variability for example, fundamental differences in ecoregion scales.In contrast, seasonality in Q TEM is caused entirely by wind speed, not attenuated by roughness, and explains the omission of the July mode and the erroneous dominance of the November mode (Figure 1d).The difference between Q AEM and Q TEM demonstrates the influence of aerodynamic roughness included and omitted, respectively (Figures 1c and 1d).
There is little resolution dependence in total F cal < 111 km resolution (Figure 1e), without any modifications to the wind fields.The linear scaling of albedo calibrated to u s* /U h has overcome the non-linear scaling of F in all but the largest resolutions.In contrast, F TEM has a much larger response to different resolutions caused by the nonlinearity in upscaling parameters (Figure 1f; cf.Chappell, Webb, Hennen, Zender, et al., 2023).Of greater concern for modeling precision is that the seasonality of F TEM is bimodal (dominated by wind) in contrast to F cal (interacting wind and roughness).The seasonality of albedo-based F cal < 111 km resolution shows little resolution dependence during June-August months and to a lesser extent also during December-February.In contrast, F cal > 111 km is more resolution dependent during March-April and to a lesser extent during October-November.Since resolution dependency is much less evident in u s* than in Q AEM , the resolution dependency is isolated (Equation 2) to grain-scale u *ts H.As resolution increases the influence of u *ts H increases (Equation 2) becoming increasingly resolution dependent.Furthermore, the sequence of Q AEM by resolution (Figure 1c) is different to that of F cal (Figure 1e) indicating that the controlling factor responsible for resolution dependency of F cal is different from Q AEM .The Q AEM is converted to F AEM using a grain-scale empirical sandblasting efficiency (Equation 3) as a proportion of soil clay content.
Although provided over area (250 m pixels), relative to the large resolutions considered here, this is point support.
The sandblasting efficiency increases by nearly 3 orders of magnitude as soil clay content increases from 0% to 20% (Zender et al., 2003).Therefore, the increase in F cal at finer resolutions (0.5 and 5 km; inconsistent with Q AEM ), is a consequence of the additional incompatibility between the grain-scale, clay-driven sandblasting and its areal implementation.It is also very likely that the sequence differences in F cal and Q AEM between resolution, is caused by using native 11 km winds with the finer (0.5 and 5 km) albedo-based roughness.We investigate the spatial distribution of these resolution dependent processes next.Given the word limit, we include a summary of results from the classical approach in the supplement and only briefly refer to them below.

Regional Spatial Differences in Dust Emission With Resolution
We mapped the factors controlling F cal at two resolutions (Figure 2) displayed spatially restricted to 278 km resolution.We consider the dominant March dust mode (from Figure 1) to make differences between models clear without aggregations across other months.The March maximum u s* in North Africa is smaller than those in Central and East Asia.The 0.5 km resolution is very similar to that at 278 km across all the major dust producing regions (Figures 2a and 2b).The number of "drag days", when soil surface wind friction velocity exceeds the sediment entrainment threshold (u s* >u *ts (d)H(w) assuming for simplicity values between d = 50-125 μm are similar, we used d = 125 μm), are also quite similar between these resolutions and across regions (Figures 2c and  2d).However, notable differences in "drag days" between resolutions in Australia and particularly the Middle East and North Africa have evidently caused the resolution dependency we noted earlier in the global seasonality of Q AEM (Figure 1c).The clay content is much smaller in North Africa than other global dust source regions (Chappell, Webb, Hennen, Schepanski, et al., 2023).The clay content is similar between resolutions, indicating that the combination of the Q AEM differences and the clay differences influence the F cal differences with resolution (Figures 2e and 2f).The combination of small clay content and intermediate u s* causes March F cal to be larger in regions other than North Africa (Chappell, Webb, Hennen, Schepanski, et al., 2023).Overall, the F cal maps show very similar spatial distributions regardless of resolution (Figures 2g and 2h).However, missing F cal is evident in Australia and Central and East Asia and the largest magnitude difference occurs in North Africa and North America.We quantify F cal by region and resolution next.
The amount of F cal for the year 2022, by region and by resolution is shown in Table 1 (uncalibrated F TEM are shown in Text S4 in Supporting Information S1; Table S1 in Supporting Information S1).At 0.5 km resolution, the total F cal is 19 Tg y 1 , broadly consistent with the recent global albedo-based average (2001-2020) F cal and its uncertainty (29.1 ± 14.9 Tg y 1 ; Chappell, Webb, Hennen, Schepanski, et al., 2023;Chappell, Webb, Hennen, Zender, et al., 2023).The amount estimated here is smaller than that recent estimate because we consider only the year 2022 and because the 278 km spatial restriction excludes coastal regions (Figure 2).With resolutions ≤11 km there is little difference in total F cal (0.55 Tg y 1 ; 2.89%) which is due to the native wind speed scale of 11 km ) and approximately one third of our recent global F cal uncertainty (±14.9Tg y 1 ).Without modifying wind fields, global F cal decreased by only 2 Tg y 1 (10.5%) from 0.5 to 111 km resolution.Our results represent fully dynamic F cal that is, spatio-temporal variation in driving variables across all scales of soil and vegetation.Our F cal results are around one third of the ∼29% decrease for the smaller range of 28-110 km resolution in F TEM (Table S1 in Supporting Information S1) and those elsewhere (Feng et al., 2022).Our F TEM calculations showed 96% difference between 0.5 and 278 km which demonstrates considerably more resolution dependency than the F cal (Table S1 in Supporting Information S1).That resolution dependency of F TEM is primarily because dust emission has a fixed drag partition, controlling the wind friction velocity, which does not change with scale.Ridley et al. (2013) showed similarly large differences in their F across a smaller range of resolutions (33% from 28-200 km).However, their experiments assumed uniform soil texture and uniform surface roughness which reduced their experiments sensitivity to land surface properties (Ridley et al., 2013).Decoupling wind speed from aerodynamic roughness (changing over space and time) in that way, will introduce additional uncertainty in the parameterization used to reconcile differences in F between resolutions.
We explained elsewhere how and why North Africa and the Middle East are not persistently the largest global sources of F, and that regional F contributions shift seasonally (Chappell, Webb, Hennen, Schepanski, et al., 2023).At 0.5 km resolution, our largest F cal is from Australia followed closely by the Middle East and North Africa (Table 1).At 278 km resolution, the largest amount of global F is from the Middle East, North Africa and then Australia (Table 1).The dominance of Australia at 0.5 km is explained by our use of only a single year and interannual variability in dust emission, and because the F cal are restricted spatially using the 278 km resolution which preferentially focuses on continental interiors due to pixel edge effects (Figure 2).Australian F cal naturally occurs away from the highly vegetated coastal regions.Consequently, using this 278 km resolution spatial restriction gives preference to F cal from continental interiors and causes Australia to have the most dust in these simulations.However, the purpose of the simulations is to compare the relative differences.Australia has the greatest difference in F cal between the 0.5 and 278 km resolution of 1.95 Tg y 1 (40.16% of the total difference Table 1).The greatest resolution dependence is very likely caused by Australia having the largest differences in landscape variability at 278 km for example, with fundamental differences in ecoregion scales at 278 km.East Asia and North America have the next largest differences (∼0.91 Tg y 1 around 19%; Table 1) in F cal caused by these resolutions.

Conclusion
Our calibrated albedo-based dust emission (F cal ) magnitude, spatial distribution, and seasonality does not depend upon resolution (<111 km).Area-integrated albedo scales linearly with resolution and its relation with sheltering represents sub-grid scale heterogeneity in u s* .This straightforward scaling in a global F cal produced remarkably consistent results across resolution, with fewer parameters than F TEM , making the approach suitable for large scale dust cycle modeling and operational weather forecasting.At 111 km and particularly at 278 km resolution, linear scaling of albedo and calibration to u s* /U h behaves differently because of a fundamental difference in the landscape heterogeneity for example, ecoregion scale.
This dust emission modeling approach is largely resolution independent without adjusting wind fields, unlike previous approaches.Dust emission is not dependent solely on wind speed (U h ), but on u s* which represents aerodynamic sheltering.The long-standing focus on adjusting the wind field excludes the feedback and interaction with aerodynamic roughness changes over space-time.Consequently, attempts to improve F models using new re-gridded wind field parameterizations do not adequately account for the sub-grid scale heterogeneity in the u s* , which drive F. Hence, existing approaches to large scale modeling of dust emission retain considerable resolution dependence.
Overcoming non-linearity in scaling u s* already demonstrated elsewhere the importance of seasonally shifting global dust source regions with many other arising implications for this largely resolution independent F modeling.ESMs cycling dust with this new resolution independent modeling will very likely reduce uncertainty and correctly represent dust types and seasonality particularly those overlooked in the southern hemisphere.These ESMs will then be able to tackle previously unresolved issues for example, Southern Ocean productivity

Geophysical Research Letters
10.1029/2023GL106540 associated with quartz and iron hypotheses from dust sources.Similarly, properly coupling albedo-based dust cycling with energy-driven ESMs offers new opportunities for feedbacks and interactions.Furthermore, dust aerosol modeling accounting for sub-grid scale heterogeneity with this approach is expected to provide more accurate air quality forecasts than classical approaches.
Future improvements in F modeling must overcome the remaining non-linearity of grain-scale parameterizations being re-gridded over large scales.Remaining resolution dependency in Q seasonality was evident in regional differences in "drag days" and in F cal by grain-scale sandblasting and is also likely to occur in the soil moisture function H(w) and wind speed U h .The long-standing u *ts and sandblasting parameterizations are responsible for the remaining F resolution dependence important for dust cycle models.Until the u *ts and sandblasting are parameterized non-linearly (and ideally dynamically) over area, F modeling will remain poorly constrained and resolution dependent.

Figure 1 .
Figure 1.Seasonality of monthly 2022 dust emission for different resolutions (0.5 km, 5 km, 11 km, 28 km, 55 km, 111 km, and 278 km) using albedo-based F cal modeling (left column) and uncalibrated F TEM (right column) using upscaled soil surface wind friction velocity normalized by wind speed (u s* /U h ; (a) and (b), sediment flux Q; (c) and (d) and dust emission (F; e and f).

Figure 2 .
Figure 2.For March 2022, using 0.5 km (left column) and 278 km (right column) model resolution and spatially restricting to the latter, the soil surface wind friction velocity (u s* m s 1 ; a and b), the number of "drag days" when soil surface wind friction velocity exceeds the sediment entrainment threshold (u s* >u *ts (d)H(w) fixing d = 125 μm; c and d), soil clay content restricted to a 20% maximum (e) and (f) and calibrated albedo-based dust emission (F cal kg m 2 y 1 ; g and h).The driving variables are reprojected by forcing the images at 0.5 km to be computed at 278 km.

Table 1
Calibrated Total Emitted Mass (Tg y 1 ) During 2022 for ModelResolutions (0.5 km, 5 km, 11 km, 28 km, 55 km, 111 km, and 278 km)of the Main Global Dust Source Regions Using Upscaled Albedo Calibrated to Wind Friction Velocity larger than those fine resolutions.At model resolutions >11 km the total F cal doubles to the largest difference of 4.86 Tg y 1 ( 25.6%) when comparing 0.5-278 km resolution.However, 4.86 Tg y 1 is similar to the F cal model uncertainty (±3.8 Tg y 1 being