Assessment of natural and anthropogenic aerosol air pollution in the Middle East using MERRA-2, CAMS data assimilation products, and high-resolution WRF-Chem model simulations

. Modern-Era Retrospective analysis for Research and Applications v.2 (MERRA-2), Copernicus Atmosphere Monitoring Service Operational Analysis (CAMS-OA) data assimilation products, and a regional Weather Research and Forecasting model (10 km resolution) coupled with Chemistry (WRF-Chem) were used to evaluate natural and anthropogenic aerosol air pollution in the ME during 2015-2016. Satellite and ground-based AOD observations, as well as in situ Particulate Matter (PM) measurements for 2016, were used for validation. 5 WRF-Chem code was modiﬁed to correct the calculation of dust gravitational settling and aerosol optical properties. The dust emission in WRF-Chem is calibrated to ﬁt Aerosol Optical Depth (AOD) and aerosol volume size distributions obtained from Aerosol Robotic Network (AERONET) observations. MERRA-2 was used to construct WRF-Chem initial and boundary conditions both for meteorology and chemical/aerosol species. SO 2 emissions in WRF-Chem are based on the novel NASA SO 2 emission dataset that reveals unaccounted sources over the ME. 10 Although aerosol ﬁelds in WRF-Chem and assimilation products are quite consistent, WRF-Chem, due to its higher spatial resolution and better SO 2 emissions, is preferable for analysis of regional air-quality over the ME. The WRF-Chem’s PM background concentrations exceed the World Health Organization (WHO) guidelines over the entire ME. The major contributor to PM ( ≈ 75–95%) is mineral dust. In the ME urban centers and near oil recovery ﬁelds, non-dust aerosols (primarily sulfate) contribute up to 26% into PM2.5. The contribution of sea salt into PM can rich up to 5%. The contribution of organic matter 15 into PM prevails over black carbon. to discuss the spatial patterns of aerosol pollution and partial contributions from natural and anthropogenic aerosols into PM. Thus, in this study, we found that MERRA-2 and CAMS-OA assimilation products, as well as WRF-Chem output despite some intrinsic uncertainties, could be successfully used for evaluation the air-quality over the Arabian Peninsula. All products show the dominant contribution of mineral dust into PM. However, in the Arabian coastal areas where SO 2 emissions are high, both contributions of sulfate and sea salt could be signiﬁcant. The developed WRF-Chem modeling framework can be used to 650 simulate other pollutants like NO 2 and O 3 . The results of the current research could serve as the basis for an improved air-quality forecast system that interactively calculates high-resolution radiative, dynamical, atmospheric chemistry and aerosol processes, driven by natural and anthropogenic emissions. This system will be especially valuable for the prediction of extreme pollution events. It will also improve understanding of the impact of anthropogenic and natural pollution on air quality and human health in the ME region. on MERRA-2 reanalysis for a WRF-Chem simulation by interpolating chemical species mixing ratios deﬁned on the MERRA-2 grid to the WRF-Chem grid for initial conditions and boundary conditions. In the case of initial conditions, interpolated values are written to each node of the WRF-Chem grid. In the case of boundary conditions, only boundary nodes are 670 affected.

Aver. period WHO US-EPA EC KSA-PME In Cuevas et al. (2014), atmospheric mineral dust from the MACC reanalysis has been evaluated over the MENA region for 85 2007-2008 using satellite and ground-based observations. MERRA-2 and CAMS-OA are global and have a relatively low spatial resolution (in comparison with the regional models), which diminishes their ability to resolve fine-scale regional spatial features. They improve the aerosol total column loadings through the assimilation of observed AOD but are not capable of assimilating the aerosol vertical structure and chemical composition. Like any other model, MERRA-2 and CAMS-OA use emission inventories of anthropogenic pollutants that may be outdated and incomplete, especially in the rapidly developing 90 parts of the world, like the ME region (McLinden et al., 2016). Here we improve the latest inventories of anthropogenic emissions in WRF-Chem using the novel SO 2 emissions data set (Liu et al., 2018).
Thus in this study, we test aerosol outputs from MERRA-2, CAMS-OA, and WRF-Chem over the ME, against satellite, ground-based AOD observations, and in situ PM 2.5 and PM 10 measurements, and evaluate air-quality over the ME focusing on the following science questions: 1. How accurately do WRF-Chem, MERRA-2, and CAMS-OA capture the abundance of dust aerosol, its volume size, and spatial distributions over the ME, in comparison with AERONET and satellite observations?
2. How accurately do WRF-Chem, MERRA-2, and CAMS-OA capture PM surface concentrations compared with in situ measurements?
3. What are the contributions of dust, sea salt, sulfate, black carbon, and organic matter in PM surface concentrations? 100 4. What is the overall impact of PM pollution on air quality over the ME region and in the most significant ME urban centers?
The paper is organized as follows: Section 2 describes the observational datasets used in this study. Section 3 briefly describes data assimilation products. In Section 4, the WRF-Chem model setup is described. In Section 5, a comparison of the capabilities of WRF-Chem, MERRA-2, and CAMS-OA to simulate dust aerosol abundance over the ME is presented. PM pollution maps 105 and PM levels in major urban centers of the ME obtained from the WRF-Chem simulation are also discussed. Conclusions are presented in Section 6.

Observational datasets
To evaluate the data assimilation products and WRF-Chem output, we use MODIS AOD retrievals, ground-based Aerosol Robotic Network (AERONET) AOD observations, and aerosol volume size distribution retrievals, as well as in situ measure-110 ments of PM surface concentrations.

AERONET
AERONET comprises more than 1000 CIMEL and PREDE robotic sunphotometers (made in France and Japan, respectively) which measure direct sun and sky radiances at eight wavelengths (340,380,440,500,670,870,940, and 1020 nm) every 15 minutes during daylight time (Holben et al., 1998). In 2012 we established the KAUST Campus site, which is currently the only 115 operational AERONET site in Saudi Arabia. For this study we have chosen three AERONET sites (KAUST Campus, Mezaira, and Sede Boker, see Fig. 1) that routinely collected data in 2015-2016 and are located within our domain. We primarily utilized level 2.0 (cloud screened and quality assured) AERONET AOD data but used level 1.5 (cloud screened) data when level 2.0 data were not available. To facilitate comparison with the model output the 550 nm AOD is calculated using Angstrom exponent according to the following relation: 5 https://doi.org/10.5194/acp-2020-17 Preprint. Discussion started: 5 February 2020 c Author(s) 2020. CC BY 4.0 License.
where α is the Angstrom exponent, τ λ is the optical thickness at wavelength λ, and τ λ0 is the optical thickness at the reference wavelength λ 0 . From here forward, we will presume that AOD is given or calculated at 550 nm.
In addition to direct observations of AOD, the AERONET retrieval algorithm provides column integrated Aerosol Volume Size Distribution (AVSD) dV/dlnr (µm 3 /µm 2 ) on 22 logarithmically equidistant discrete points in the range of radii between 125 0.05 and 15 µm (Dubovik and King, 2000). We use these retrievals to evaluate the AVSDs produced by WRF-Chem, CAMS-OA, and MERRA-2.

MODIS
MODIS instruments onboard the NASA Terra and Aqua satellites provide aerosol properties over both land and ocean with near-daily global coverage. The high surface albedo over the desert surfaces complicates the AOD retrievals (Shi et al., 2011).

130
The standard MODIS AOD aerosol product combines two retrieval algorithms: 1) the MODIS dark-target (DT) algorithm (Kaufman et al., 1997) is used over the ocean and dark areas with sufficient vegetation, 2) the Deep Blue (DB) algorithm is used over bright desert surfaces of the Sahara and the ME. The uncertainties of AOD obtained with the DB algorithm are ≈25-30% (Hsu et al., 2006). From this combined product (MODIS-DB&DT v6.1) we use AOD at 550 nm level 3 data from the daily dataset at 1 • ×1 • spatial resolution, downloaded from https://giovanni.gsfc.nasa.gov (Acker and Leptoukh, 2007). 135 Recently, a new MODIS AOD product became available that was obtained using the Multi-Angle Implementation of Atmospheric Correction (MAIAC) algorithm (Lyapustin et al., 2018). This algorithm uses time series analysis and image processing to derive the surface bidirectional reflectance function at fine spatial resolution. MAIAC uses empirically tuned, spatially varying, aerosol properties derived from the AERONET climatology, and provides AOD at 550 nm with 1 km spatial resolution over land globally. We include the new MAIAC product (version 6, level 2) in the comparison between simulated and retrieved 140 AODs. MERRA-2 and CAMS-OA assimilate satellite and ground-based observations to provide aerosol abundance and air-quality data globally. In contrast, WRF-Chem is a free-running model and does not assimilate observations. Here, we specifically 155 evaluate these products against observations over the ME, and compare them with the WRF-Chem output.

MERRA-2
MERRA-2 (https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2) provides meteorological and atmospheric composition fields on 0.625 • ×0.5 • latitude-longitude grid and 72 terrain-following hybrid σ − p model layers (Randles et al., 2017;. The pressure at the model top equals 0.01 hPa. MERRA-2 uses the Goddard Earth Observing System, version 5 (GEOS-160 5) atmospheric model (Rienecker et al., 2008), which is interactively coupled to the Goddard Global Ozone Chemistry Aerosol Radiation and Transport (GOCART) model (Chin et al., 2002) (i.e., it takes into account the effects of aerosols on radiation and model dynamics). This model simulates dust and sea salt in five size bins (see Tab. 2), SO 2 , sulfate, organic and black carbon (hydrophobic and hydrophilic), O 3 , CO, dimethyl sulfide DM S), and methane sulfonic acid (M SA). The dust density is 2600 kg/m 3 for all sizes. Dust and sea salt emissions are calculated in the model, depending on the near-surface wind.

165
The dust source function is taken from Ginoux et al. (2001). For anthropogenic emissions, MERRA-2 employs the EDGAR-4.2 (Janssens-Maenhout et al., 2013) emission inventory available on a 0.1 • ×0.1 • grid. MERRA-2 assimilates AOD at 550 nm from the AVHRR (Heidinger et al., 2014) over the oceans, AOD retrievals from the MISR over bright surfaces (Kahn et al., 2005), as well as specially processed MODIS observations (but not the standard MODIS-DB&DT aerosol product) and AERONET to constrain the atmospheric aerosols.   1995) with the source function adopted from Ginoux et al. (2001). SO 2 oxidation into sulfate aerosol is parameterized using a prescribed latitude-dependent e-folding timescale ranging from 3 days at the equator to 8 days at the poles. The anthropogenic emissions for the chemical species are taken from the MACCity inventory (Granier et al., 2011), which is available on a 0.5 • ×0.5 • grid and covers the period 1960-2010. CAMS-OA assimilates MODIS observations.  To calculate fine-resolution PM and sulfate fields, we use the Weather Research and Forecasting (WRF) model  coupled with chemistry (WRF-Chem v3.7.1) (Grell et al., 2005). The WRF-Chem is used for prediction and simulation of weather, air quality, and dust storms, accounting for the aerosol effect on radiation. WRF-Chem can be configured with one of the few gas-phase chemical mechanisms, photolysis, and aerosols parameterization models. WRF-Chem has been widely used for air quality simulations in different parts of the globe: East Asia (Wang et al., 2010), US (Kim et al., 2006;190 Chuang et al., 2011), Europe (Forkel et al., 2012;Ritter et al., 2013), South America (Archer-Nicholls et al., 2015) and Middle East (Parajuli et al., 2019).
To reduce the clock-time of our two-year calculations, we simulated each month of the 2015-2016 period separately. Each simulation starts from the last week of the previous month. This time is considered a spin-up and is excluded from postprocessing. The simulation domain, shown in Fig. 1, is centered at 28 • N, 42 • E, and a 10 km×10 km horizontal grid (450×450 195 grid nodes) is employed. The vertical grid comprises 50 vertical levels with enhanced resolution closer to the ground comprising 11 model levels within the near-surface 1-km layer. The model top boundary is set at 50 hPa.
To improve the representation of the meteorological fields, we apply spectral nudging (Miguez-Macho et al., 2004) above the planetary boundary layer (PBL) (>5.0 km) to horizontal wind components (U and V ) toward the MERRA-2 wind field. The nudging coefficient for U and V is set to be 0.0001 s −1 . We nudge modes with wavelengths larger than 450 km. This allows 200 us to keep the large-scale motions close to reanalysis, and leave the resolved small-scale, high-frequency features unaffected.
The aerosol/chemistry initial conditions and boundary conditions (IC&BC) are calculated using MERRA-2 output by means of the newly developed Merra2BC interpolation utility (see Appendix A1). To be consistent with aerosol/chemistry IC&BC, we also define the meteorological IC&BC using MERRA-2 output (see Appendix A1).
The following set of physical parameterizations was used in WRF-Chem runs. The Unified Noah land surface model 205 (sf_surface_physics=2) and the Revised MM5 Monin-Obukhov scheme (sf_sfclay_physics=1) are chosen to represent land surface processes and surface layer physics. The Yonsei University scheme is chosen for PBL parameterization (bl_pbl_physics=1).
The WRF single moment microphysics scheme (mp_physics=4) is used for the treatment of cloud microphysics. The New Grell scheme (cu_physics=5) is used for cumulus parameterization. The Rapid Radiative Transfer Model (RRTMG) for both short-wave (ra_sw_physics=4) and long-wave (ra_lw_physics=4) radiation is used for radiative transfer calculations. Only the aerosol direct radiative effect is accounted for. More details on the physical parameterizations used can be found at http://www2.mmm.ucar.edu/wrf/users/phys_references.html.

Gas-phase chemistry and aerosols
To calculate the atmospheric chemistry within WRF-Chem, we employ the Regional Atmospheric Chemistry Mechanism (RACM, chem_opt=301) (Stockwell et al., 1997) containing 77 species and 237 reactions, which include 23 photolysis re-215 actions. It is embedded into WRF-Chem using the Kinetic PreProcessor (KPP) (Damian et al., 2002). The role of KPP is to integrate the system of stiff nonlinear ordinary differential equations, which represents the specified set of chemical reactions.
The photolysis rates are calculated on-line according to Madronich (1987) (phot_opt=1). Similar to MERRA-2, the GOCART chemistry module is used to calculate SO 2 to sulfate oxidation (Chin et al., 2002) by the hydroxide radical OH whose abundance is interactively simulated by RACM.

220
Here we use the novel OMI-HTAP SO 2 emission dataset (Liu et al., 2018) based on the combination of distributed SO 2 emissions from residential and transportation sectors, taken from the HTAP-2.2 inventory (Janssens-Maenhout et al., 2015) with the catalogue of the strong (>30 kt/year) SO 2 point emissions (Fioletov et al., 2016)

built using satellite observations by
Ozone Monitoring Instrument (OMI) (Levelt et al., 2006;Li et al., 2013). The catalogue contains more than 500 point sources of industrial origin, some of which are not present in the widely used EDGAR-4.2 and HTAP-2.2 emission datasets. For 225 example, 14 previously unaccounted SO 2 point emissions located in the ME (mostly in the Arabian Gulf) were detected, most of them are related to oil and gas industry. OMI-HTAP divides SO 2 emissions into surface and elevated ones. We distribute the surface SO 2 emissions with a constant mixing ratio in the 0-1000 m layer, and elevated emissions in 120-1000 m layer.
All other constituents (PM, black and organic carbon, etc.), including SO 2 shipping emissions, are taken from the HTAP-2.2 inventory and are treated as surface emissions.

230
To calculate aerosols we employ the GOCART (Chin et al., 2002) aerosol model (chem_opt=301). It is the same microphysical model as that used in MERRA-2 (see Sec. 3.1). Dust and sea salt size distributions in WRF-Chem are approximated by the same five dust and sea-salt size bins as those in MERRA-2 (Tab. 2). However, only the last four "salt" bins in Tab. 2 are used in WRF-Chem, as the first bin appears to be very poorly populated. Dust density is assumed to be 2500 kg/m 3 for the first dustbin and 2650 kg/m 3 for 2-5 dustbins. Emission of sea salt is calculated according to Gong (2003). Dust emission from the 235 surface is calculated using the GOCART emission scheme (Ginoux et al., 2001) (dust_opt=1). Dust emission mass flux, F p (µg m −2 s −1 ) in each dustbin p=1,2,...,5 is defined by the relation: where, C has the dimension of (µg s 2 m −5 ) and is a spatially uniform factor which controls the magnitude of dust emission flux; S is the spatially nonuniform topographic source function (Ginoux et al., 2001) that characterizes the spatial distribution 9 https://doi.org/10.5194/acp-2020-17 Preprint. Discussion started: 5 February 2020 c Author(s) 2020. CC BY 4.0 License. of dust emissions; u 10m is the horizontal wind speed at 10 m; u t is the threshold velocity, which depends on particle size and surface wetness; s p is a fraction of mass emitted into dustbin p, s p = 1.
To avoid natural dust emission in urban areas, we use the built-in WRF-Chem the U.S. Geological Survey (USGS) 24category land-use data set (Anderson, 1976). We modify the source function S using the following expression: where U RBAN _M ASK is the USGS "Urban and Built-up Land" mask field. It has the sense of a fraction of urban area 245 in a grid-cell and ranges from 0 to 1. Grid cells with U RBAN _M ASK=1 do not produce natural dust emissions. We do not account for anthropogenic dust emissions within cities, and therefore potentially underestimate urban dust pollution.
As in our previous studies (Kalenderski et al., 2013;Jish Prakash et al., 2015;Anisimov et al., 2017), we tune dust emissions to fit the daily average AOD from the AERONET stations located within the domain. For this purpose, the parameter C from Eq.
(2) has been adjusted to achieve the best agreement between simulated and observed AOD at KAUST Campus, Mezaira, Obtained during test runs, C value of 0.5 is kept constant in all subsequent production runs. We also tune s p from Eq. (2) to better reproduce the AVSDs provided by AERONET inversion algorithm. This tuning and the comparisons of AOD and AVSDs from the assimilation products and WRF-Chem simulations are discussed in detail below.

WRF-Chem code modification
We have corrected the source code of the WRF-Chem v3.7.1 with GOCART aerosol module in several places. These corrections were implemented in the WRF-Chem v4.1.3 official release and will be described in the forthcoming technical publication.
Here we briefly discuss the introduced changes and their effects.
Firstly, the diagnostic output of PM concentrations was corrected, because contributions of the individual dust and sea salt 260 bins were incorrectly calculated. Therefore, PM 2.5 surface concentrations were erroneously underestimated while PM 10 -were overestimated.
Secondly, we found that the contribution of fine dust particles with radii <0.46 µm was omitted in the calculation of AOD.
This led to an overestimation of the dust emission flux because we force the simulated AOD to match the AERONET observations.

265
Thirdly, we fixed the dust and sea salt gravitational settling subroutine, since initially, the calculations of mass fluxes of settling particles did not account for changes in air density. Due to this error, the total mass of dust and sea salt aerosols increased, violating mass conservation.

Regional climate and circulation 270
The ME is one of the hottest and driest regions on the Earth. Summer in the ME is long and hot with little precipitation.
Precipitation mainly occurs in the south-west of the Arabian Peninsula. Winter is mild, with rainfall being mostly associated with cold fronts and cyclones propagating from the Eastern Mediterranean (Climate.com, 2018). Emission and transport of dust are driven by winds. Emission and deposition of dust are also sensitive to soil moisture and precipitation (Furman, 2003;Shao, 2008;Yu et al., 2015).  Over northeast Africa in winter (see Fig. 2a), the strong pressure gradient between the Red Sea trough and the stationary high-pressure system over Egypt predominantly generates moderate north-easterly winds (up to 10 m/s). Therefore in winter, dust storms occur more frequently in the west of the Arabian Peninsula. Over the Central and Eastern Arabian Peninsula and 280 11 https://doi.org/10.5194/acp-2020-17 Preprint. Discussion started: 5 February 2020 c Author(s) 2020. CC BY 4.0 License. In summer (see Fig. 2b the high-pressure system over the eastern Mediterranean and low-pressure system over the Arabian Gulf promote moderate north-northwesterly winds known as Shamal (Yu et al., 2016;Hamidi et al., 2013), which dominate over the central part of the Arabian peninsula. Shamal is the primary driver of dust storm events over this area (Yu et al., 2016;285 Shao, 2001;Middleton, 1986;Goudie and Middleton, 2006;Notaro et al., 2015). Shamal brings dust to the Arabian Gulf, north, and central part of Saudi Arabia, from the Tigris-Euphrates basin of Syria and Iraq (Anisimov et al., 2018). 3, suggest that WRF-Chem captures the magnitude and spatial distribution of the 10 m wind. Thus, we conclude that WRF-

310
Chem with the selected set of physical parameterizations satisfactorily simulates both the large-and meso-scale atmospheric processes in the ME. Table 3. Pearson correlation coefficient R and root mean square difference RM SD (m/s) for the seasonally averaged 2015-2016 wind components U and V at 10 m.

AOD
In this section, we evaluate the ability of WRF-Chem, CAMS-OA, and MERRA-2 to reproduce the aerosol content in the atmosphere accurately. This content is characterized by AOD. In the ME, mineral dust contribution to the total AOD is dominant 315 (≈87%) (Kalenderski and Stenchikov, 2016;Osipov et al., 2015). The treatment of optically active dust within the model is therefore vitally important. AOD is calculated based on aerosol concentrations and aerosol optical properties, which depend upon aerosol size distribution. We, therefore, evaluate how well WRF-Chem and assimilation products reproduce aerosol size distribution.

Aerosol size distributions 320
Dust particles are emitted into the lower atmospheric layer with some predominant size distribution (Kok, 2011). Emitted dust is processed by the atmosphere to produce the atmospheric dust size distribution that is retrieved by the AERONET inversion algorithm (Dubovik and King, 2000) and reported as column integrated AVSD. Strictly speaking, AERONET AVSD incorporates contributions from all types of aerosols. But because dust dominates all other aerosols in the ME, we have to tune the dust emission parameters in the first place.  AERONET sites (see Fig. 1), since only these sites have information on AVSDs during the 2015-2016 period. Because dust bins are coarse, especially in the sub-micron range, model and assimilation products struggle to correctly reproduce the fine mode of the AERONET AVSD (see Fig. 4). The volume size distributions from the model and assimilation products demonstrate pronounced seasonal variability with the increased amount of dust in the atmosphere during spring and summer. Since the KAUST Campus and Mezaira sites are located in the vicinity of the strong dust sources, the coarse mode at these sites is 340 more pronounced than at the Sede Boker site, which is farther from the strong dust emission sources.
Both MERRA-2 and WRF-Chem use the GOCART aerosol scheme with the same five dustbins, and they approximate the shape of the AERONET AVSD relatively well. CAMS-OA uses only three dustbins (see Tab. 2) and fails to reproduce the AERONET AVSD even qualitatively. It overestimates the volume of particles with radii of 0.55-0.9 µm and underestimates  The fine mode in the AERONET AVSD is more pronounced at the KAUST Campus in comparison with the other AERONET sites due to its proximity to strong SO 2 sources located along the west coast of Saudi Arabia. This proximity leads to a higher contribution of fine sulfate particles to the fine mode. The smaller volume of fine particles in the WRF-Chem and MERRA-2 simulated dust AVSD (see Fig. 4) is in part because the sulfate contribution is not shown. Sea salt particles/droplets are 350 relatively large and mostly contribute to the coarse mode. Figure 5 shows the contributions of dust, sea salt, and sulfate aerosols into the AVSD at the KAUST Campus AERONET site in WRF-Chem simulation averaged for two summer seasons (JJA) of 2015-2016. In WRF-Chem, sulfate aerosol is computed using a bulk approach. However, for calculating of aerosol optical properties, it is assumed that sulfate aerosol is described by two log-normal modes: nuclei and accumulation. According to WRF-Chem source code, the nuclei mode radii µ nuc =0.005 µm 355 and geometric width σ nuc =1.7, the accumulation mode radii µ acc =0.035 µm and geometric width σ acc =2.0. The nuclei mode comprises 25% of the sulfate aerosol mass, and accumulation mode -75%. It is assumed that sulfate aerosol density is 1800 kg/m 3 and sea salt density is 2200 kg/m 3 . Figure 5 demonstrates that the contribution of the sulfate nuclei mode in the aerosol volume is almost negligible, while the sulfate accumulation mode adds in the volume of aerosol particles with radii <1 µm.
The contribution of the sea salt aerosol into AVSD in WRF-Chem simulations is very little.

Comparison of spatial AOD distributions
We also examine how well MERRA-2, CAMS-OA, MAIAC, and WRF-Chem reproduce spatial patterns and seasonal vari-  MAIAC AOD, which has the lowest correlation (0.608) and highest RMSE (0.135). Notably, the difference in terms of R and RMSE between the two retrieval algorithms MODIS-DB&DT and MAIAC is bigger than the difference between WRF-Chem and MODIS-DB&DT. Based on the comparison of WRF-Chem AOD with the AOD from AERONET and MODIS observations, we conclude 405 that spatial and temporal WRF-Chem's AOD distribution is in good agreement with the available ground-based and satellite observations.

Air-quality
To test the ability of the data assimilation products and models to characterize air-quality in the ME, we compare surface daily mean PM 2.5 and PM 10 concentrations from WRF-Chem, MERRA-2, and CAMS-OA, with daily averaged measurements 410 20 https://doi.org/10.5194/acp-2020-17 Preprint. Discussion started: 5 February 2020 c Author(s) 2020. CC BY 4.0 License.
conducted by the three AQMS, see Fig. 8 and 9. The AQMS are installed in Jeddah, Riyadh, and Dammam (Fig. 1) where DD 1,2,3 , SS 1,2 , sulf ate, BC 1,2 , OM 1,2 are surface concentration of dust in three bins, sea salt in two bins, sulfate, 430 black carbon, and organic matter (hydrophobic and hydrophilic). The size ranges of dust and sea salt bins from CAMS-OA are presented in Tab. 2.
The histograms at the right-side panels in Fig. 8 and 9 show the annual mean PM concentrations from WRF-Chem, MERRA-2, and CAMS-OA split into the dust and non-dust components. We also calculated the separate contributions of sulfate, sea salt, organic matter, and black carbon into the non-dust PM 2.5 and PM 10 , see Tab. 6 and 7, respectively. The dashed and dash-dotted concentrations, as discussed in Sec. 4.1. In Riyadh, the contribution of the non-dust component to PM 2.5 is ≈9-11% for both MERRA-2 and WRF-Chem. In CAMS-OA, the contribution of non-dust particulates to PM 2.5 in Jeddah and Dammam is ≈7-11%, and the contribution of sea salt is little. According to Tab. 6, in all considered cities, the contribution of black carbon to PM 2.5 is not significant in WRF-Chem, CAMS-OA, and MERRA-2. In MERRA-2, the contribution of organic matter to PM 2.5 is more substantial (but still minor) in comparison with WRF-Chem and CAMS-OA.        Abbreviations of the aerosols' names correspond to those given in Sec. 5.3. µg/m 3 , correspondingly. The regions of high surface concentrations coincide with the main dust sources, which span from Northern Iraq to Oman, include Sudan, Egypt, Algeria, and Turkmenistan. PM concentrations in these regions exceed even the less restrictive KSA-PME air quality limit for annual mean PM 2.5 and PM 10 concentrations by more than 5 times.

495
In the entire domain, the max, min, and mean values of the PM 2.5 /PM 10 ratio (see Fig. 10c) are 0.89, 0.24, and 0.38 respectively. As expected, the lower PM 2.5 /PM 10 ratios (0.2-0.3) are observed over the dust source regions (i.e., along the eastern Arabian peninsula, Iraq, and northern Africa), where both coarse and fine particles are generated. However, large particles can not be transported as far from source regions as small particles, due to the shorter lifetime of large particles compared with small particles with respect to deposition processes. The higher values of the PM 2.5 /PM 10 ratio are observed 500 over south-eastern Europe, Turkey, Ethiopia, and western parts of the Arabian Peninsula that are farther from the main dust sources. Figure 10d shows the sum of surface concentrations of black carbon and organic matter (OC 1 + OC 2 ) * OC mf ac + BC 1 + BC 2 ). Their max, min, and mean concentration values are 31.8, 0.2, and 1.3 µg/m 3 respectively. Their contribution to aerosol pollution over the Arabian Peninsula in WRF-Chem simulations is fairly insignificant. Figure 10e shows the surface concen- Oman. Due to prevailing northern winds, the transport of sea salt from the Mediterranean Sea to Egypt and Libya is observed. The relatively high sulfate surface concentration (see Fig. 10f) is observed in the vicinity of the strong SO 2 sources located along the west and east coast of Saudi Arabia and over the Arabian Gulf, as well as downwind from those sources. Figure   10f also denotes the locations of the AERONET stations, as in Fig. 1. The sulfate concentration at the KAUST Campus site is higher than at the Mezaira and Sede Boker AERONET sites (see Sec. 5.2.1) so it experiences more pronounced contribution of sulfate particulates into the fine mode of the AVSD (see Fig. 4a and Fig. 5).
Due to the prevailing northern winds along the Red Sea, sulfate aerosols originating from SO 2 emissions along the Red Sea Contributions of dust to PM 2.5 and PM 10 calculated as ratios of dust PM 2.5 to total PM 2.5 , and dust PM 10 to total PM 10 , are shown in Fig. 10g and 10h respectively. Due to relatively low dust surface concentrations over the eastern Mediterranean,

525
Turkey, and south-eastern Europe, the contribution of dust to PM 2.5 and PM 10 is 20-50% and 50-80%, respectively. In other areas that are closer to the dust source regions, the contribution of dust to PM is above 90%. Figure 10i shows the ratio between the concentration of sulfate aerosol with respect to the total concentration of PM 2.5 non-dust aerosols. The max, min, and mean values of this ratio are 0.76, 0.07, and 0.45, respectively. This ratio is low over the seas where sea salt is prevalent but consistently exceeds 0.6 over land. Sulfate, therefore, is the primary anthropogenic pollutant 530 over land. In the Arabian Coastal areas, central and southern parts of Saudi Arabia, and over south Iran, sulfate contributes 60-80% to the total PM 2.5 non-dust aerosols concentration. Over the other parts of the Arabian Peninsula, the northern part of Sudan, Libya, and Egypt, sulfate contributes approximately 40-60% to total PM 2.5 non-dust aerosols concentration.

Air-pollution in urban centers
To evaluate the air-quality in the ME's major cities, we calculate for their locations the average 2015-2016 daily PM 2.5 and 535 PM 10 surface concentrations, their 90th percentiles, and we also calculate the contribution of the dust and non-dust components into PM (see Fig. 11). We calculate the number of days during the 2015-2016 period when the daily PM 2.5 and PM 10 surface concentrations exceed the US-EPA air-quality limit of 35 µg/m 3 and 150 µg/m 3 respectively. Figure 11 shows that the annually-averaged PM 10 and PM 2.5 exceed the WHO air-quality guidelines 2-20 and 2-12 times, respectively in all major cities of the ME, except Ankara. The KSA-PME air-quality limit for annual mean PM 10 is exceeded 540 by up to 5 times, and by up to 8 times for PM 2.5 . Due to the lack of strong dust sources nearby, air-quality conditions in the cities in the eastern Mediterranean are more favorable when compared with those in the Arabian Peninsula. In these cities, the air-pollution shifts from natural to anthropogenic, as the contribution of non-dust aerosols to PM 2.5 increases up to 40%, in In the scope of this study, we conducted advanced two-year high-resolution WRF-Chem simulations. The WRF-Chem code was corrected to better describe the aerosol effects, and a new capability of using MERRA-2 output for boundary and initial conditions has been developed. To improve the calculation of sulfate aerosols, the most accu-560 rate SO 2 emission dataset retrieved from OMI observations using the "top-down" approach was implemented in WRF-Chem.
The contribution of natural dust, sea salt, and anthropogenic aerosols into the PM was estimated. We found that the three-bin approximation in CAMS-OA is not enough to correctly represent the aerosol size distribution, and MERRA-2 overestimates sea salt and underestimates sulfate concentrations. The air pollution in the major Middle Eastern cities is evaluated.
We evaluate the PM air-pollution over the ME during the 2015-2016 period using the regional WRF-Chem model v3.7.1,

565
MERRA-2 and CAMS-OA assimilation products, and satellite and ground-based AOD observations, as well as in situ PM 2.5 and PM 10 surface concentration measurements available for 2016.
The regional WRF-Chem model has an advantage of higher spatial resolution ( in Dammam in comparison with Jeddah. In Dammam the KSA-PME limit for daily averaged PM 10 of 340 µg/m 3 is more frequently exceeded than in Jeddah, especially during the summer period. Annually averaged MODON measurements are 8-11 times higher than the WHO guideline of 20 µg/m 3 and 2-3 times higher than the KSA-PME limit of 80 µg/m 3 .
The capability of WRF-Chem, MERRA-2, and CAMS-OA in reproducing the ME air-quality is tested against AQMS measurements. In Jeddah and Riyadh WRF-Chem and MERRA-2 are able to reproduce the PM 10 measurements quite well, but are higher than in WRF-Chem and MERRA-2 capturing PM 2.5 observations better than other products in Jeddah and Riyadh.
We use WRF-Chem output to conduct the PM composition analysis. We found that the annual average PM 2.5 /PM 10 ratio over the ME is 0.38. It decreases to 0.2-0.3 over the major dust source regions, i.e. along the eastern Arabian peninsula, Iraq, and northern Africa. In most parts of the ME, dust is the major contributor to PM, but in the eastern Mediterranean and Turkey contribution of the dust component to PM 2.5 and PM 10 decreases to 20-50% and 50-80%, respectively.

620
Sulfate aerosol contributes to PM 2.5 in the areas where strong SO 2 sources are present, i.e., in the west and east coasts of Saudi Arabia and over the Arabian Gulf. In these areas sulfate surface concentration reaches 8-11 µg/m 3 , while the "clean" background level is 2-4 µg/m 3 . High sulfate content along the west coast of Saudi Arabia is consistent with the increased volume of the fine mode in the KAUST Campus AERONET AVSD in comparison with AVSDs from other sites. Sulfate is the major non-dust pollutant over the ME. Sulfate aerosols contribute 60-80 % to the total PM 2.5 non-dust aerosols over the 625 western and eastern Arabian coasts, over the central and southern parts of Saudi Arabia, and over the southern Iran. Over the other parts of Arabian Peninsula, northern Sudan, Libya, and Egypt, sulfate contributes approximately 40-60 % to the total PM 2.5 non-dust aerosol concentration.
The analysis of the annually averaged PM 2.5 and PM 10 surface concentrations in the ME major cities conducted using WRF-Chem output shows a very high level of PM pollution. In Dammam, Abu Dhabi, Doha, Kuwait City, and Baghdad, the 630 90th percentile of PM 10 and PM 2.5 annual mean surface concentrations exceed 500-750 and 150-230 µg/m 3 respectively, which is above the KSA-PME air-quality limit. In the eastern Mediterranean, dust concentration drops, and the contribution of non-dust aerosols to PM 2.5 increases up to 25-40%. In the cities located in the Arabian Peninsula contribution of the non-dust component to PM 2.5 is 6-26%, which limits the emission control on air-quality. In the eastern Mediterranean cities during the 2015-2016 period, the daily mean surface PM concentrations exceed the US-EPA air quality daily mean limit 5-75 days for 635 PM 10 and 7-208 days for PM 2.5 . In the ME cities over the Arabian peninsula, Iraq, and Iran, the US-EPA air-quality daily mean limit is exceeded 95-626 days for PM 10 and 230-684 days for PM 2.5 .
In Jeddah and Dammam, WRF-Chem and MERRA-2 show similar contributions of the non-dust component to PM 2.5 (25-33% in Jeddah and 10-14% in Dammam). In MERRA-2, however, sea salt is a major non-dust contributor to PM 2.5 , while in WRF-Chem it is sulfate. In CAMS-OA contribution of the non-dust particulates to PM 2.5 in Jeddah and Dammam is ≈7-11%.

640
In Riyadh, the contribution of the non-dust component to PM 2.5 is ≈9-11% for both MERRA-2 and WRF-Chem.
In Jeddah, Riyadh, and Dammam, the contribution of black carbon to PM 2.5 and PM 10 is not significant in WRF-Chem and both assimilation products. In MERRA-2, the contribution of organic matter to PM 2.5 is more substantial in comparison with WRF-Chem and CAMS-OA. However, in CAMS-OA, PM 10 has more organic matter than in WRF-Chem and MERRA-2.
We also observe the relative increase of organic matter in PM 2.5 (except WRF-Chem in some cases) and PM 10 in Jeddah and Thus, in this study, we found that MERRA-2 and CAMS-OA assimilation products, as well as WRF-Chem output despite some intrinsic uncertainties, could be successfully used for evaluation the air-quality over the Arabian Peninsula. All products show the dominant contribution of mineral dust into PM. However, in the Arabian coastal areas where SO 2 emissions are high, both contributions of sulfate and sea salt could be significant. The developed WRF-Chem modeling framework can be used to 650 simulate other pollutants like N O 2 and O 3 . The results of the current research could serve as the basis for an improved airquality forecast system that interactively calculates high-resolution radiative, dynamical, atmospheric chemistry and aerosol processes, driven by natural and anthropogenic emissions. This system will be especially valuable for the prediction of extreme pollution events. It will also improve understanding of the impact of anthropogenic and natural pollution on air quality and human health in the ME region.
The full MERRA-2 reanalysis dataset including aerosol fields is publicly available online (see "Code and data availability" section). Depending on the requirements, one or both of the following aerosol and gaseous collections need to be downloaded:

675
inst3_3d_aer_N v -gaseous and aerosol mixing ratios (kg/kg) and inst3_3d_chm_N v -Carbon monoxide and Ozone mixing ratios (kg/kg). In addition to downloaded mixing ratios, pressure thickness DELP and surface pressure PS fields also need to be downloaded. Spatial coverage of the MERRA-2 files should include the area of the WRF-Chem simulation domain. The time span of the downloaded files should match with the start and duration of the WRF-Chem simulation. For more information regarding MERRA-2 files specification please refer to Bosilovich et al. (2016).

A1.1 Mapping chemical species between MERRA-2 and WRF-Chem
The Merra2BC input file conf ig.py contains multiplication factors to convert MERRA-2 mixing ratios of gases given in kg/kg to ppmv. Aerosols are converted from kg/kg to µg/kg. When using the GOCART aerosol module in the WRF-Chem simulation, all MERRA-2 aerosols and gases are matched with those from WRF-Chem. To convert MERRA-2 aerosol mixing ratios given in kg/kg into µg/kg, multiply by a factor of 10 9 . In the case of gases, multiply MERRA-2 mixing ratios by a ratio 685 of molar masses M air /M gas multiplied by 10 6 to convert kg/kg to ppmv, where M gas and M air are the corresponding molar masses. If another aerosol module is chosen in WRF-Chem, then different multiplication factors should be used.

A1.2 Typical workflow
Below are the steps describing how to work with the Merra2BC utility: 1. Run real.exe, which will produce the initial wrf input_d01 and boundary conditions wrf bdy_d01 files required by (c) Path to the downloaded MERRA-2 collection files 5. Program real.exe sets default boundary and initial conditions for some chemical species. Merra2BC adds interpolated values to the existing ones and it may cause incorrect concentration values. To avoid this, run script "zero_f ileds.py", which will zero the required fields 700 6. Run script "main.py", which will perform the interpolation; as a result, files wrf input_d01, wrf bdy_d01 will be updated by the values interpolated from MERRA-2 7. Modify WRF-Chem namelist.input file at section &chem: set have_bcs_chem = .true. to activate updated boundary conditions from MERRA-2 and, if it is needed, chem_in_opt = 1 to activate updated initial conditions; 8. Run wrf.exe program. For the usage of the Merra2BC interpolator the following python modules need to be installed: -netcdf4: https://github.com/Unidata/netcdf4-python scipy: https://github.com/scipy/scipy

A2 Statistics
We calculated the following statistical parameters to quantify the level of agreement between estimations and observations: Pearson correlation coefficient (R): Root mean square error (RM SE): Mean bias (BIAS): where