Application of WRF-CAMx over West Asia, Part I: Meteorological and Air Quality Model Evaluation

Highlights

What are the main findings?

• The WRF-CAMx system reasonably reproduces meteorology and air quality over West Asia, with good performance for temperature, wind, PM_2.5, and AOD.
• Model limitations include underestimation of humidity and cloud effects in winter and larger errors in trace gas concentrations.

What are the implications of the main findings?

• Improving emissions inventories and chemical mechanisms is essential to reduce biases in trace gases across the region.
• Refinements in model physics, particularly cloud and humidity representation, are critical for better seasonal performance.

Abstract

Air pollution poses significant risks to public health, ecosystems, and regional economies, particularly in rapidly developing regions. Despite its importance, the Middle East remains relatively understudied in regional air quality, with limited evaluations of pollutant transport and model performance. This study applies the WRF (Weather Research and Forecasting) model coupled with the CAMx (Comprehensive Air Quality Model with Extensions) model to simulate meteorology and air quality over West Asia, with a focus on the United Arab Emirates (UAE). Six representative months are analyzed, including three winter periods (January 2018, 2020, 2022) and three summer periods (June 2017, 2019, 2021). WRF shows good agreement with observations, reproducing near-surface temperature with an index of agreement (IOA) between 0.90 and 1.00 and generally low wind speed (MB < ±0.5 m s⁻¹) and wind direction biases (MB < ±0.5), although cloud-radiative forcing is underestimated during winter. CAMx reproduces PM_2.5 concentrations with moderate-to-high correlations (r = 0.44–0.65) and low bias, while AOD and O₃ column concentration show larger uncertainties. Satellite-based evaluation indicates good performance for NO₂ and CO column abundances but larger discrepancies for HCHO and SO₂, particularly during summer. Overall, the results demonstrate that the WRF-CAMx modeling system provides a reliable framework for regional air quality simulations over West Asia, while highlighting uncertainties associated with emissions, atmospheric chemistry, and satellite retrieval products.

1. Introduction

Air pollution poses a significant environmental challenge in multiple regions worldwide, such as West Asia, which covers several fast-growing countries, including the United Arab Emirates (UAE) [1,2]. The UAE has experienced substantial growth in commerce, tourism, and industry. Its unique geographical setting (characterized by arid desert conditions, intense solar radiation, and the presence of the Persian Gulf and Gulf of Oman) creates complex atmospheric dynamics that influence air quality. Rapid urbanization with increased vehicular traffic, industrial activities, and large-scale infrastructure development has significantly elevated concerns about air pollution in the country [3].

In addition to local emissions, the UAE is affected by transboundary pollution and natural sources such as dust storms, which further complicate air quality management. Particulate matter from wind-blown dust is an important source of air pollutants in the region that affects not only the air quality locally [4,5] but also the air quality in various regions depending on the atmospheric conditions [6] through long-range transport.

Several studies have advanced our understanding of dust storms in West Asia, revealing their variability, transport mechanisms, and modeling challenges. Ref. [7] analyzed extreme dust events over the Arabian Peninsula from 2003 to 2017 using satellite data, while ref. [6] explored seasonal variability and transport pathways of dust aerosols. Ref. [8] examined frontal dust storms and their atmospheric circulation patterns. Modeling studies by reference [9] compared dust emission schemes (Air Force Weather Agency scheme, AFWA, reference [10] versus reference [11]) in WRF-Chem, finding significant differences in particle size distributions of the dust emissions and PM₁₀ concentrations. Refs. [12,13] both used WRF-Chem to study dust storm impacts on the Arabian Peninsula and Red Sea, demonstrating reductions in surface solar radiation and effects on regional climate and biogeochemical cycles through nutrient deposition. Despite these advancements, significant gaps persist in accurately predicting West Asian dust storms. While ref. [14] improved dust modeling by incorporating the West Asia Source Function (WASF) into WRF-Chem, showing improvements over reference [15], unidentified dust sources and uncertainties in soil moisture data continue to affect model outputs. Ref. [16] found that increased spatial resolution alone does not improve dust predictions, highlighting the need to identify additional performance factors.

Air quality modeling in West Asia faces several challenges that limit its accuracy. One major gap is the limited availability of comprehensive, high-quality, and long-term air quality observational data across the region [7], and, consequently, there is no systematic evaluation of meteorological variables and atmospheric composition, which are essential for accurately simulating atmospheric chemistry and pollutant transport. Many models struggle to accurately represent dust emissions and transport due to uncertainties in identifying natural dust sources and complex atmospheric circulation patterns. Additionally, there is a lack of comprehensive sector-specific inventories for anthropogenic emissions, leading to additional uncertainties. High-resolution modeling remains challenging due to computational constraints and uncertainties in local-scale meteorological processes and emissions. Moreover, few studies systematically validate models against ground-based and satellite observations or conduct multi-model intercomparisons to assess model reliability. Finally, air quality models are underutilized in policy-driven research, limiting their role in assessing emission control measures and climate adaptation strategies.

The coupled WRF-CAMx modeling system has been applied to investigate air quality, pollutant transport, and aerosol processes across a variety of environments. Previous studies have used WRF-CAMx to evaluate ozone, particulate matter, and trace gas concentrations in North America [17,18], Europe [19,20], and East Asia [21,22], demonstrating its capability to reproduce regional meteorology and atmospheric chemistry while identifying key uncertainties associated with emissions, chemical mechanisms, and meteorological inputs. These applications have shown that the accuracy of air-quality simulations strongly depends on the quality of the meteorological fields provided by WRF, as well as on the representation of emissions and boundary conditions. Consequently, comprehensive evaluations of both meteorological and air-quality performance remain an essential step before using the modeling system for scientific or regulatory applications.

This work is presented in two parts. Part I (this manuscript) focuses on applying and validating the Weather Research and Forecasting (WRF) and Comprehensive Air Quality Model (CAMx) for West Asia. Part II (this manuscript) focuses on ozone (O₃) formation regimes and the processes controlling O₃ and PM_2.5 over the United Arab Emirates (UAE). The goal of Part I is to assess the models’ ability to simulate meteorological conditions and air pollutant distributions through comparisons with available ground-based and satellite observations, highlighting the main gaps and limitations of the current modeling framework.

2. Materials and Methods

The WRF-CAMx simulations are performed using double-nested domains, with the coarse domains covering most of West Asia and the nested domains covering the UAE. WRF and CAMx are configured with different spatial resolutions to optimize computational efficiency and meet the modeling objectives. Figure 1 shows the WRF domains (dashed lines) at 27 km and 9 km resolutions, along with the CAMx domains (solid lines) at 20 km and 10 km resolution; the elevation is represented by the color scale. The WRF domains are intentionally larger than the corresponding CAMx domains to avoid using meteorological results near the model boundaries, where numerical artifacts and boundary condition influences can negatively affect the CAMx results. The 27 km WRF outer domain provides boundary conditions for the 9 km nest, which better resolves wind fields and boundary layer dynamics. CAMx uses a 20 km outer domain to better resolve emission sources and chemical gradients, which typically show sharper spatial variations than meteorological fields, while the 10 km inner domain is appropriate for regional air quality modeling given the available emission inventory resolution. The WRF model is a widely used numerical weather prediction system that incorporates various physical parameterizations to simulate atmospheric processes [23], simulating meteorological variables such as wind fields, temperature, pressure, and humidity that serve as input for CAMx (WRF 27 km outputs are processed for CAMx 20 km and WRF 9 km outputs for CAMx 10 km). CAMx simulates air quality processes, including the transport, dispersion, chemical transformation, and deposition of atmospheric pollutants [24].

In this study, version 4.3.1 of the WRF model is used. Both domains use a vertical resolution of 35 layers. Table 1 summarizes its physics and assimilation options. The simulations are performed using the Rapid Radiative Transfer Model for General circulation model (RRTMG) schemes for shortwave and longwave radiation, which incorporates cloud and aerosol interactions and an accurate treatment of the gaseous and aerosol absorption [25]. For microphysics, the Morrison two-moment scheme is used, which predicts both the mass and number concentrations of key hydrometeor species, including cloud water, rain, cloud ice, snow, and graupel [26]. For the surface layer, the option Pleim–Xiu Scheme is selected, providing the representation of fluxes of momentum, heat, and moisture between the land surface and the lower atmosphere [27]. Further, this scheme dynamically adjusts soil moisture and temperature using a force-restore approach, improving the representation of surface energy fluxes and boundary layer evolution [28,29]. The Asymmetric Convective Model version 2 (ACM2) scheme is used for boundary layer processes. This hybrid scheme combines local and non-local closure to represent both small-scale diffusion and large-scale eddy transport, allowing for upward and downward mixing [30]. For the cumulus option, the Tiedtke scheme of reference [31] is chosen. This option accounts for deep, shallow, and mid-level convection, making it suitable for tropical and mid-latitude convection processes.

The meteorological initial and boundary conditions for the WRF simulations are from the Global Forecast System model from the National Centers for Environmental Prediction (NCEP GFS) with a spatial resolution of 0.25°. Land-use characteristics are from the United States Geological Survey (USGS) land-use dataset. To improve the accuracy of the meteorological variables, Four-Dimensional Data Assimilation (FDDA) is applied using the NCEP final operational global analysis (NCEP GDAS/FNL) with a resolution of 0.25° and deep soil nudging using Meteorological Aerodrome Reports (METARs) observations from the Meteorological Assimilation Data Ingest System (MADIS) system [32]. The assimilation is performed every six hours for all model layers outside the Planetary Boundary Layer (PBL), with nudging coefficients for wind speed (guv), temperature (gt), and humidity (gq) as detailed in Table 1. WRF uses fixed time steps determined according to the grid spacing of each domain (120 s for the 27 km domain and 40 s for the 9 km domain). Additionally, the simulation is initialized 10 days prior to the analysis period to allow for model spin-up.

Table 1. WRF and CAMx model options.

WRF Parameter	WRF Option
Shortwave radiation	RRTMG shortwave [25]
Longwave radiation	Updated RRTMG scheme [25]
Cloud microphysics	Morrison 2-moments Scheme [26]
Surface layer	Pleim-Xiu [27]
Land surface	Pleim-Xiu Land Surface Model [28,29,33]
Boundary layer	ACM2 PBL [30]
Cumulus clouds	Tiedtke scheme [31,34]
FDDA	guv = 0.0003, gt = 0.0003, gq = 0.00001
Deep soil nudging	Pleim & Gilliam [32]
CAMx Parameter	CAMx Option
Chemistry mechanism	CB6r4 + DMS + CF2 [35]
Aerosol treatment	Coarse/fine CF 2-mode model (CF2)
Inorganic PM chem	ISORROPIA [36,37]
Organic PM chem	SOAP2.2 [38]
Horizontal advection	PPM [39]
Chemistry solver	EBI [40]
Dry deposition model	WESELY89 [41,42]
Inline Ix emissions	Yes
Super stepping option	Yes

RRTMG: Rapid Radiative Transfer Model for General circulation model, ACM2: Asymmetric Convection Model 2 Scheme, PBL: Planetary Boundary Layer, CB6r4: Carbon Bond v6, DMS: including dimethyl sulfide oxidation reactions, CF2: coarse/fine model, ISORROPIA: Improved State Of Reversible Reactions Reacting Organic particles in the atmosphere aerosol thermodynamic equilibrium model, SOAP2.2: Secondary Organic Aerosol Partitioning, PPM: piecewise parabolic method, EBI: Euler backward iterative, Inline Ix emissions: oceanic inorganic iodine emissions, Super stepping: option that allows the model to use large time-steps possible, keeping the solution numerically stable.

Table 1. WRF and CAMx model options.

WRF Parameter	WRF Option
Shortwave radiation	RRTMG shortwave [25]
Longwave radiation	Updated RRTMG scheme [25]
Cloud microphysics	Morrison 2-moments Scheme [26]
Surface layer	Pleim-Xiu [27]
Land surface	Pleim-Xiu Land Surface Model [28,29,33]
Boundary layer	ACM2 PBL [30]
Cumulus clouds	Tiedtke scheme [31,34]
FDDA	guv = 0.0003, gt = 0.0003, gq = 0.00001
Deep soil nudging	Pleim & Gilliam [32]
CAMx Parameter	CAMx Option
Chemistry mechanism	CB6r4 + DMS + CF2 [35]
Aerosol treatment	Coarse/fine CF 2-mode model (CF2)
Inorganic PM chem	ISORROPIA [36,37]
Organic PM chem	SOAP2.2 [38]
Horizontal advection	PPM [39]
Chemistry solver	EBI [40]
Dry deposition model	WESELY89 [41,42]
Inline Ix emissions	Yes
Super stepping option	Yes

RRTMG: Rapid Radiative Transfer Model for General circulation model, ACM2: Asymmetric Convection Model 2 Scheme, PBL: Planetary Boundary Layer, CB6r4: Carbon Bond v6, DMS: including dimethyl sulfide oxidation reactions, CF2: coarse/fine model, ISORROPIA: Improved State Of Reversible Reactions Reacting Organic particles in the atmosphere aerosol thermodynamic equilibrium model, SOAP2.2: Secondary Organic Aerosol Partitioning, PPM: piecewise parabolic method, EBI: Euler backward iterative, Inline Ix emissions: oceanic inorganic iodine emissions, Super stepping: option that allows the model to use large time-steps possible, keeping the solution numerically stable.

The CAMx model (version 7.20) is used. Both domains use a vertical resolution of 22 levels. The first 15 layers correspond to the first 15 levels of WRF, and the remaining 7 layers are mapped to WRF model layers of 16 to 32 with spacing gradually increasing up to WRF level 32.

Table 1 also provides a summary of the physics, chemistry, and other configurations used in the CAMx simulations. For horizontal advection, CAMx uses the Piecewise Parabolic Method (PPM) [39]. The dry deposition is modeled using WESELY89, a resistance-based approach that accounts for the deposition of gases [42] and aerosols [41]. For the chemical mechanism, the Carbon Bond v6 (CB6r4) scheme [43] is used, including dimethyl sulfide (DMS) oxidation reactions [44]. This mechanism includes 233 reactions among 87 species (62 state gases and 25 radicals). The coarse and fine aerosol (CF2) scheme divides the size distribution into two static modes. Primary species are modeled as fine and/or coarse particles, while all secondary (chemically formed) species are modeled as fine particles only. Partitioning of inorganic aerosol constituents (sulfate, nitrate, chloride, ammonium, and sodium) between the gas and aerosol phases is performed using ISORROPIA v1.7 [36,37]. For secondary organic aerosol (SOA) chemistry, the SOAP2.2 mechanism [38] is applied. These mechanisms are solved using the Euler Backward Iterative (EBI) method [40], which provides a balance of computational speed and accuracy. The model is run using the super stepping option that allows the model to select the largest grid-specific driving time steps possible [24].

Anthropogenic emissions are generated using R (version 4.2.0) and Python (version 3.11.0) scripts, in combination with the EmissV R-package [45], the emissions from the Emissions Database for Global Atmospheric Research EDGAR 5.0 [46] (native resolution of 0.1° × 0.1°) are allocated to the CB6r4 mechanism Volatile Organic Compounds (VOCs) species using data from the Speciation Tool [47]. All EDGAR 5.0 available emission sectors were processed and applied as surface emissions within the CAMx modeling system. There are many uncertainties regarding the total estimated emissions in EDGAR for the region of West Asia; for example, the uncertainty for emissions is 69.4% for NO_x, 108.1% for CO, 47.7% for SO₂, 133.9% for VOCs, 56.5% for PM_2.5, 104.3% for PM₁₀ [48]. Initial model applications using the original EDGAR emissions showed systematic biases in simulated pollutant concentrations relative to observations, as summarized in Table S1. The scaling factors were empirically selected based on the sign and approximate magnitude of the normalized mean biases identified during the initial model evaluation. To partially account for these uncertainties, empirically derived constant scaling factors were applied to the model-ready emissions as first-order adjustments. Specifically, NO_x emissions were reduced by 20%, while emissions of CO, SO₂, VOCs, PM_2.5, and PM₁₀ were increased by 120%, 35%, 102%, 300%, and 250%, respectively. These adjustments are intended to partially account for uncertainties in emission magnitude, rather than to represent optimized or universally applicable correction factors, and they do not address uncertainties related to sectoral representation or spatial allocation in the inventory.

Dust and oceanic emissions are generated using CAMx preprocessors. The CAMx wind-blown dust preprocessor (WBDUST) version 2.0 is used. Initial simulations showed substantial overprediction of particulate matter concentrations associated with dust emissions, particularly during strong wind events. Because WBDUST dust emissions are highly sensitive to friction velocity and surface characteristics, relatively small uncertainties in wind speed, soil properties, and land surface datasets can lead to very large uncertainties in simulated dust emissions over arid regions. To partially account for these uncertainties and reduce systematic PM overprediction, PM₁₀ dust emissions were reduced by 90% and PM_2.5 dust emissions by 97% as first-order sensitivity adjustments. Oceanic emissions are generated using the CAMx oceanic preprocessor version 4.3, and the CAMx options for inline emissions of inorganic iodine are activated. Additionally, ozone column and photolysis rates are generated using O3MAP version 3.1 and the Tropospheric Ultraviolet and Visible (TUV) radiation model (version 4.8).

Boundary conditions and initial conditions for CAMx/CB6r4 species are extracted from Community Atmosphere Model with Chemistry (CAM-Chem) outputs for simulations covering 2017–2020 and from Whole Atmosphere Community Climate Model (WACCM) outputs from simulations covering 2021–2022. Adjustments to ozone boundary conditions are applied based on an initial model evaluation at a rural monitoring station in the region. For winter months (January 2019 and January 2020), a 20% reduction is applied, while summer boundary conditions are increased by 80% in June 2017, 20% in June 2018, 40% in June 2020, and 60% in June 2021. These adjustments are determined after reducing local emissions to zero to isolate the influence of lateral boundary conditions.

Model performance is evaluated using available observational datasets.

Table 2. Observational datasets for model evaluation.

Dataset	Type	Variables	Spatial Resolution	Temporal Resolution
METAR	Surface	T2, RH2, WS10, WD10	127 sites	hourly
U.S. Embassy	Surface	PM2.5	6 sites	hourly
AERONET	Surface	AOD	5 sites	daily
GPCP v3.2	Satellite	RAIN	0.5º	monthly
TRMM	Satellite	RAIN	0.25º	monthly
CERES	Satellite	GLW, GSW, LWCF, SWCF, OLR, SWDOWN	1º	monthly
MODIS	Satellite	CDNC, CCN, CF, COT, CWP, AOD	1º	daily
MAIAC	Satellite	AOD	1 km	daily
MOPITT	Satellite	CO	1º	monthly
OMI	Satellite	NO2, O3, SO2, and HCHO	0.25º	daily
AIRS v7.0	Satellite	O3, CF, OLR	1º	monthly

T2: temperature at 2 m, RH: relative humidity, WS: wind speed, WD: wind direction, PM_2.5: particulate matter ≤ 2.5 µm, AOD: aerosol optical depth, RAIN: precipitation, GLW: long-wave radiation, GSW: short-wave radiation, LWCF: long-wave cloud forcing, SWCF: short-wave cloud forcing, OLR: outgoing long-wave radiation, SWDOWN: downward long-wave radiation, CDNC: cloud droplet number concentration, CCN: cloud condensation nuclei, CF: cloud fraction, COT: cloud optical thickness, CWP: cloud water path, NO₂: nitrogen dioxide column, CO: carbon monoxide column, O₃: ozone column, SO₂: sulfur dioxide column, HCHO: formaldehyde column.

summarizes the datasets used to assess meteorological variables from WRF and atmospheric composition from CAMx. Hourly data from 127 surface stations, obtained from the Meteorological Aerodrome Reports [49], are used to evaluate meteorological variables, including temperature at 2 m (T2), specific humidity at 2 m (Q2), wind speed at 10 m (WS10), and wind direction at 10 m (WD10). Additionally, aerosol optical depth (AOD) is evaluated using ground-based remote sensing data from the Aerosol Robotic Network [50]. AOD is a dimensionless measure of the effect of the absorbing or scattering of sunlight by the aerosols in the atmosphere. Of the 28 AERONET stations located within the CAMx 20 km simulation domain, five have AOD data available for the simulated periods. The satellite products are used to validate the model covering various meteorological and atmospheric composition variables. Precipitation predictions are obtained from the Global Precipitation Climatology Project (GPCP) v3.2 [51] and Tropical Rainfall Measuring Mission-TRMM satellite [52], which provide the long-term global precipitation data.

Radiation flux components are evaluated using products from the Clouds and the Earth’s Radiant Energy System (CERES) [53,54], which offer detailed information on longwave (GLW) and shortwave (GSW) radiation at surface levels, the Longwave Cloud Forcing (LWCF) and Shortwave Cloud Forcing (SWCF), and Outgoing Longwave Radiation (OLR) and Downward Shortwave Radiation at the Surface (SWDOWN). Additionally, cloud parameters such as cloud droplet number concentration (CDNC), cloud condensation nuclei (CCN), cloud optical thickness (COT), cloud water path (CWP), and cloud fraction (CL) are compared with data from the Moderate Resolution Imaging Spectroradiometer (MODIS) [55]. CF and OLR are compared with Atmospheric Infrared Sounder (AIRS) v7.0 [56].

AOD is compared with MODIS and Multi-Angle Implementation of Atmospheric Correction (MAIAC) [57]. To evaluate atmospheric composition, column pollutant concentrations are analyzed using data from the Measurement of Pollution in the Troposphere [58], which provides carbon monoxide (CO) column predictions, and the Ozone Monitoring Instrument (OMI) [59], which provides column concentrations of nitrogen dioxide (NO₂), Ozone (O₃), sulfur dioxide (SO₂), and formaldehyde (HCHO). Finally, AIRS is also used to evaluate O₃ column. These satellite datasets serve as critical references for assessing CAMx model performance, since there is a lack of data available for the study domain.

To evaluate the performance of the WRF and CAMx models, statistical analysis is conducted using criteria from the literature and implemented through the R-package eva3dm [60]. For evaluation using ground-based data, observed data is spatially and temporally paired with model time series. In the case of the evaluation using satellite data, model results and satellite observations are first averaged monthly, followed by interpolation using the bilinear method of the satellite data to the model grid. To ensure reliable comparisons, only data points located at least five grid points away from the model boundary are considered. The evaluation includes the calculation of common statistical metrics such as mean error (ME), mean bias (MB) of the absolute value, index of agreement (IOA), Root Mean Square Error (RMSE), and correlation coefficient (r), as well as modified normalized mean bias (NMB) and modified normalized mean error (NME), both computed using an absolute-value denominator to ensure consistent normalization for variables that may take negative values.

The evaluation criteria for model performance vary depending on the variable, simulation goals, observational variability, and measurement errors. For meteorological variables such as temperature at 2 m (T2), specific humidity at 2 m (Q2), wind speed (WS), and wind direction (WD), criteria from [61,62] are applied for simple terrain, while [63] is referenced for complex terrain. For precipitation, the evaluation follows the recommendations of [64].

Unlike meteorological variables, there are no standardized evaluation criteria for AOD in the literature. Most existing guidance comes from studies conducted in the U.S., including the early work by [64,65] and the more comprehensive assessment by [66]. For the study region, several publications exist [67,68,69,70,71,72] but they do not provide a consistent set of evaluation benchmarks. Thus, we review these studies to identify commonly used criteria that can guide the evaluation of AOD for the region. One common metric in these studies is the use of r2 values, which typically range from 0.5 to 0.8. Bias metrics such as MB and NMB are commonly used, with acceptable thresholds between 20% and 40% [70,73,74]. However, ref. [70] note that global biases often compensate for regional errors. Ref. [72] report a mean error (ME) of 0.3 for polluted regions (AOD ≥ 1) and 0.05 for less polluted areas (AOD < 0.5). In this study, AOD evaluation is based on r, NMB, and NME, with threshold values derived from previous literature. The selected thresholds are comparable to those for PM_2.5 proposed by [75], where NMB ranges between 10% and 30% and NME between 35% and 50%. For the PM_2.5 from satellite, we are using thresholds based on the previously mentioned literature: values equal to or higher than 0.7 indicate good model performance, while values equal to or higher than 0.4 indicate acceptable performance, which are stricter than traditional evaluation against surface stations in [75].

For column O₃ abundances, the evaluation follows the criteria from [75], included in

Table 3. Criteria and thresholds used for the WRF and CAMx models’ evaluation.

Model	Variable	Criteria	Benchmarks
			Simple Terrain 1	Complex Terrain 2
WRF	T2	IOA	≥0.8
		MB	≤±0.5 °C	<±1.0 °C
		ME	≤2 °C	<3.0 °C
	Q2	IOA	≥0.6
		MB	≤±1 g/kg
		ME	≤2 g/kg
	WS	MB	≤±0.5 m/s	<±1.5 m/s
		RMSE	≤2 m/s	<2.5 m/s
	WD	MB	≤±10 deg
		ME	≤30 deg	<55º
	RAIN 3	NMB	≤±30%
	AOD 4		Good performance	Acceptable performance
CAMx		r	>0.8	>0.6
		NMB	<±20%	<±40%
		NME	<30%	<50%
	PM2.5 5	r	>0.7	>0.4
		NMB	<±10%	<±30%
		NME	<35%	<50%
	O3 5	r	>0.75	>0.5
		NMB	<±5%	<±15%
		NME	<15%	<25%
	Col. CO 5	NMB		<±15%
	Col. NO2 5	NMB		<±30%
	Col. SO2 5	NMB		<±30%
	Col. HCHO 5	NMB		<±50%

IOA: index of agreement, MB: mean bias, ME: mean error, RMSE: Root Mean Square Error, NMB: normalized mean bias, r: correlation coefficient, and NME: normalized mean error. 1—criteria from references [61,62] for simple terrain. 2—criteria from reference [63] for complex terrain. 3—criteria from reference [65]. 4—based on the literature [67,68,69,70,71,73]. 5—criteria from reference [75].

. For the column abundances of CO, NO₂, SO₂ and HCHO only the NMB metric is considered, with acceptable threshold values of ±15%, ±30%, ±30%, and ±50%, respectively, to assess the degree of underprediction or overprediction [75].

The evaluation conducted in this study is affected by the limited availability of observational datasets over West Asia. Surface observations for air pollutants remain sparse across much of the region, particularly for PM_2.5, aerosol properties, and trace gas measurements, relying partially on satellite retrieval products that contain additional uncertainties related to retrieval assumptions, cloud contamination, spatial representativeness, and vertical sensitivity. Although multiple independent datasets were combined to support the model evaluation, additional surface observations and aerosol profile measurements would be essential for a more in-depth assessment of model performance and atmospheric processes.

3. Results and Discussion

Meteorological variables from the WRF simulations for domains d01 (27 km) and d02 (9 km), as well as chemical results from the CAMx simulations for domains d01 (20 km) and d02 (10 km), are presented in this section. The analysis covers six simulated cases: three winter cases (January 2018, January 2020, and January 2022) and three summer cases (June 2017, June 2019, and June 2021). The analyzed simulation periods were selected based on the availability of meteorological and air quality observations and their consistency with the typical seasonal conditions identified in broader simulations conducted for 2017–2022. Although the selected months do not represent the full interannual variability of the region or all extreme dust and pollution episodes, they provide representative winter and summer conditions suitable for evaluating the performance of the WRF-CAMx modeling system under distinct meteorological and chemical environments over West Asia.

3.1. WRF Evaluation

Table 4. Evaluation of WRF meteorological variables.

Season	Variable	Domain	n	Obs	IOA	MB	ME	RMSE	NMB
Winter	T2	27 km in d01	214,223	32.8	0.96	−0.6	2.2	2.8	−1.8
		27 km in d02	67,307	34.6	0.92	−0.6	2.3	2.9	−1.7
		9 km in d02	67,307	34.6	0.91	−0.5	2.5	3.2	−1.5
	Q2	27 km in d01	191,904	9.7	0.88	−0.7	3.1	4.4	−6.8
		27 km in d02	64,574	14.4	0.84	−1.1	4.3	5.9	−7.7
		9 km in d02	64,574	14.4	0.83	−1.0	4.5	6.2	−6.7
	WS	27 km in d01	190,189	4.2	0.73	0.1	1.7	2.3	2.3
		27 km in d02	60,029	4.5	0.78	−0.3	1.6	2.2	−7.0
		9 km in d02	60,029	4.5	0.78	−0.4	1.6	2.2	−8.8
	WD	27 km in d01	209,363	189.2	0.93	5.3	49.7	67.5	2.8
		27 km in d02	65,100	192.3	0.93	3.8	46.8	65.3	2.0
		9 km in d02	65,100	192.3	0.93	3.8	46	64.8	2.0
	RAIN GPCP	27 km in d01	34,347	0.7	0.84	−0.1	0.4	0.8	−13.8
		27 km in d02	4332	1.2	0.84	−0.2	0.7	1.2	−15.6
		9 km in d02	57,132	1.1	0.81	−0.1	0.7	1.3	−11.2
	RAIN TRMM	27 km in d01	11,449	0.2	0.82	0.0	0.2	0.4	−8.3
		27 km in d02	1444	0.2	0.35	−0.1	0.1	0.3	−73.8
		9 km in d02	19,044	0.2	0.52	−0.1	0.1	0.3	−66.8
Summer	T2	27 km in d01	217,939	12.5	0.97	−0.1	2.0	2.7	−0.8
		27 km in d02	70,639	16.8	0.96	0.4	1.8	2.4	2.5
		9 km in d02	70,639	16.8	0.96	0.4	1.9	2.4	2.6
	Q2	27 km in d01	203,855	5.9	0.87	−0.4	1.6	2.1	−6.3
		27 km in d02	68,102	7.7	0.86	−0.6	1.7	2.3	−7.9
		9 km in d02	68,102	7.7	0.85	−0.6	1.8	2.3	−7.7
	WS	27 km in d01	182,992	3.7	0.65	0.2	1.6	2.6	4.9
		27 km in d02	60,858	4.0	0.79	0.1	1.5	2.0	2.0
		9 km in d02	60,858	4.0	0.80	0.0	1.5	2.0	−0.4
	WD	27 km in d01	212,735	170.4	0.92	6.4	53.8	72.1	3.8
		27 km in d02	68,496	188.6	0.93	5.0	47.9	66.4	2.6
		9 km in d02	68,496	188.6	0.94	6.3	45.5	63.4	3.4
	RAIN GPCP	27 km in d01	34,347	0.27	0.83	−0.10	0.19	0.74	−37.53
		27 km in d02	4332	0.04	0.50	−0.03	0.04	0.08	−62.83
		9 km in d02	57,132	0.04	0.41	−0.02	0.05	0.13	−53.08
	RAIN TRMM	27 km in d01	22,898	0.31	0.78	−0.07	0.23	1.06	−23.13
		27 km in d02	2888	0.01	0.34	0.00	0.02	0.03	−5.71
		9 km in d02	38,088	0.02	0.12	0.00	0.03	0.14	18.49

1—“27 km in d01” and “27 km in d02” indicate that the 27 km WRF simulation results are evaluated against all observations within the d01 and d02 domains, respectively (see Figure 1). Additionally, “9 km in d02” indicates that the 9 km simulation results are evaluated against all observations in the d02 domain. 2—shaded values can be compared to literature-based thresholds; unshaded values have no such reference. 3—winter and summer consider all the observations for each season in the calculation of the evaluation metrics.

shows the evaluation of WRF results for all domains using surface stations from METAR for T2, Q2, WS10, and WD10 and precipitation from satellite from Global Precipitation Climatology Project (GPCP) and Tropical Rainfall Measuring Mission (TRMM), however TRMM is available only until 2019. The evaluation of meteorological variables shows different degrees of performance across different seasons. T2 shows good agreement with observations, with an IOA ranging from 0.9 to 1.0. It exhibits a moderate seasonal bias, with an underprediction in winter (MB ≈ −0.5 °C) and an overprediction in summer (MB ≈ 0.4 °C), while maintaining moderate errors (ME between 1.71 and 2.52 °C), that are considered acceptable for complex terrain. Good performance is mainly attributed to the use of deep soil nudging, which reduces the MB and RMSE while increasing IOA. Q2 shows good agreement in terms of IOA (0.83–0.88), but a consistent underprediction (MB ranging from −0.37 to −1.1 g kg⁻¹) that exceeds the recommended benchmarks. Q2 errors are moderate in summer but much larger in winter, particularly in the 9 km WRF domain, where ME ranges from 1.57 to 4.48 g kg⁻¹ (outside the benchmarks). WS10 presents a low bias (MB between −0.4 and 0.18 m s⁻¹) and moderate errors (RMSE between 2 and 2.6 m s⁻¹) that are close to the benchmarks for complex terrain. WD10 also has a low bias (MB of 3.8 to 6.4°) but moderately large errors (ME ranging from 45 to 53°), that are within the benchmark for complex terrain. The systematic underprediction of Q2 during winter is observed at both coastal and inland stations despite the generally good performance of T2, WS, and WD. This suggests that the bias is not primarily associated with errors in the large-scale meteorological conditions or atmospheric circulation. Instead, the dry bias likely reflects limitations in the representation of processes controlling near-surface moisture availability and moisture exchange.

Precipitation shows a moderate negative bias in winter (NMB between −15% and −11%) and a significant negative bias in summer (NMB between −37% and −53%) when compared to GPCP data. However, when evaluated against TRMM data, precipitation bias is lower in summer (NMB between −23% and 18%) and significant in winter (NMB between −8% and 73%). Notably, TRMM data is only available until 2019, and comparisons with TRMM for winter and summer are similar to the GPCP comparisons for January 2018 and June 2017, respectively.

The results from the 27 km and 9 km domains can be intercompared by analyzing the rows in Table 1 labeled “27 km in d02” and “9 km in d02.” Statistically, the evaluation metrics indicate that the model performances at both resolutions are quite similar. However, some differences emerge, particularly in precipitation. The 9 km simulation shows a reduced underprediction of precipitation compared to the 27 km simulation, suggesting that the higher resolution captures precipitation patterns more accurately. Additionally, during summer, the 9 km simulation exhibits a slight improvement in WS10 performance.

Table 5. Evaluation of WRF radiation and cloud variables.

Season	Satellite	Variable	n	Obs	r	MB	NMB	RMSE	NME
Winter	CERES	GLW	34,347	308.5	0.97	−8.6	−2.8	14.5	3.4
		GSW	34,347	140.6	0.92	23.9	17.0	29.4	17.6
		LWCF	34,347	9.4	0.79	−3.9	−41.8	5.0	44.2
		SWCF	34,347	−12.6	0.70	−6.4	−50.8	9.1	54.2
		OLR	34,347	266.6	0.98	−1.5	−0.6	4.6	1.3
		SWDOWN	34,347	177.6	0.98	29.0	16.3	30.2	16.3
	MODIS	CF	34,347	0.3	0.22	−0.1	−37.5	0.2	50.7
		CWP	31,552	72.4	−0.02	16.9	23.4	197	151.6
	AIRS	CF	34,347	0.3	0.12	−0.1	−20.7	0.2	50.4
		OLR	34,347	270.8	0.96	−5.7	−2.1	8.7	2.5
Summer	CERES	GLW	34,347	391.6	0.92	−14.1	−3.6	19.5	3.9
		GSW	34,347	249.3	0.65	34.9	14.0	43.0	14.4
		LWCF	34,347	11.6	0.81	−1.0	−8.6	6.6	44.4
		SWCF	34,347	−8.2	0.81	−0.5	−6.7	7.2	57.1
		OLR	34,347	307.7	0.95	−10.9	−3.6	12.9	3.7
		SWDOWN	34,347	317.6	0.95	36.8	11.6	38.3	11.6
	MODIS	CF	34,347	0.2	−0.24	0.1	41.6	0.4	159.2
		CWP	30,240	62.9	−0.01	8.6	13.7	196.6	144.8
	AIRS	CF	34,347	0.2	−0.32	0.0	18.5	0.3	104.1
		OLR	34,347	315.7	0.93	−19.0	−6.0	21.1	6.1

1—winter and summer consider all the observations for each season to the calculation of the evaluation metrics.

evaluates the WRF results for radiation variables (GLW, GSW, LWCF, SWCF, OLR, and SWDOWN) and cloud-related variables (CF, CWP) with data retrieved from satellite observations. The radiation variables generally show good performance, with correlation coefficients ranging from 0.65 to 0.98 and low NMB (lower than 15%) and NME (lower than 30%) for most variables, these results are comparable to the findings on the literature for this type of evaluation [76,77,78,79]. However, LWCF and SWCF, which are linked to cloud processes, exhibit lower correlations and higher NMB and NME compared to other radiation variables for the winter session. This is an indication that the radiative effect of clouds is underestimated in winter; during summer, similar NME indicates that the model representation has limitations on the representation of the cloud structure and radiative effect. Other cloud-related variables, CF and CWP, show the worst performance, reflecting the high uncertainties associated with model cloud parameterizations, that can be associated with the representation of high clouds [79] in the Morrison cloud microphysics scheme [26].

The seasonal surface temperature biases in the WRF simulations are linked to systematic errors in cloud–radiation interactions. During winter, a negative temperature bias (MB = −0.5 °C) is associated with substantial underpredictions in LWCF (NMB = −41%) and CF (NMB = −37%), leading to insufficient longwave trapping and enhanced radiative cooling. Although OLR is only slightly underpredicted, the large LWCF and cloud fraction biases suggest that the modeled clouds are inefficient at modulating the surface energy balance. In summer, the model exhibits a positive 2 m temperature bias (MB = 0.4 °C), which can be attributed to underpredictions in cloud forcing from LWCF (NMB = −8.6%) and overprediction of SWCF (NMB = 6.7%). Notably, the model overpredicts cloud fraction relative to MODIS and AIRS retrievals, yet still reflects insufficient shortwave radiation, indicating that simulated clouds may be optically thin or poorly represented in the radiation scheme. These findings suggest that uncertainties in cloud microphysics and radiation parameterizations can significantly impact temperature simulations, particularly during winter.

Uncertainties in model inputs, particularly land-use and elevation datasets, can introduce significant biases in WRF simulations. Land-use classification directly affects surface energy fluxes, vegetation-driven moisture availability, and the development of the atmospheric boundary layer. These processes are critical for accurately representing near-surface temperature, wind, and humidity [80]. Elevation errors are especially problematic in regions with complex terrain, such as West Asia, where they can distort terrain-induced circulations, influence orographic precipitation patterns, and misrepresent wind flow, contributing to observed errors in wind speed and direction [8]. The arid climate and highly heterogeneous land cover further complicate the representation of land–atmosphere interactions [81,82]. Additionally, uncertainties in the meteorological initial and boundary conditions can propagate through the simulation period, affecting large-scale circulation patterns and regional dynamics.

The performance of the WRF meteorological model plays a critical role in determining the accuracy of CAMx air quality simulations, as meteorological fields govern atmospheric transport, dispersion, mixing, cloud processes, deposition, and chemical transformation of pollutants [83,84]. In this study, discrepancies in meteorological variables, particularly under stable wintertime conditions with limited atmospheric mixing, are likely to propagate into CAMx, influencing the representation of pollutant concentrations. Inaccuracies in temperature, humidity, and wind fields can lead to errors in simulating mixing layer height, horizontal and vertical transport pathways, and chemical reaction rates. These effects ultimately alter the spatial and temporal distributions of both primary and secondary pollutant formation.

3.2. CAMx Evaluation

Table 6. Evaluation of PM_2.5 and AOD against surface data.

Season	Variable	Domain	n	Obs	r	NMB	NME
Winter	PM2.5	20 km in d01	818	35.9	0.44	−25.4	44.2
		20 km in d02	396	27.2	0.55	−11.2	42.1
		10 km in d02	396	27.2	0.49	0.5	45.5
	AOD	20 km in d01	77	0.2	0.38	9.6	49.0
		20 km in d02	43	0.2	0.41	21.7	50.9
		10 km in d02	43	0.2	0.40	25.0	52.4
Summer	PM2.5	20 km in d01	852	39.0	0.51	−28.6	41.3
		20 km in d02	393	35.0	0.65	−22.7	38.9
		10 km in d02	393	35.0	0.58	−14.0	39.8
	AOD	20 km in d01	75	0.3	0.25	42.6	70.1
		20 km in d02	29	0.4	−0.13	18.0	60.9
		10 km in d02	29	0.4	−0.14	21.5	63.0

1—“20 km in d01” and “20 km in d02” indicate that the 20 km CAMx simulation results are evaluated against all observations within the CAMx d01 and d02 domains, respectively (see Figure 1). Additionally, “10 km in d02” indicates that the 10 km simulation results are evaluated against all observations in the CAMx d02 domain. 2—shaded values can be compared to literature-based thresholds; unshaded values have no such reference. 3—winter and summer consider all the observations for each season in the calculation of the evaluation metrics.

shows the evaluation of the CAMx results using surface observations for PM_2.5 at the U.S. Embassy sites and AOD from AERONET. The comparison indicates good model performance for PM_2.5 and AOD. PM_2.5 exhibits good correlation (ranging from 0.44 to 0.65) and moderately large negative bias (NMBs between −28% and 0.5%), and NMEs range from 39% to 45.4%, which fall within the acceptable to good model performance criteria outlined in Table 3. The CAMx 10 km results show an approximately 10% improvement in NMB for both cases, indicating reduced bias compared to the CAMx 20 km results. However, the 10 km resolution simulation gives slightly lower correlation values (r) and no significant change in NME. The AOD comparison reveals low correlation (r between −0.14 in summer and 0.41 in winter), NMBs ranging from 9% to 42%, and NME ranging from 49% to 52% in winter and from 60% to 70% in summer. Time series of PM_2.5 and AOD are included in the (Figure 1, Figure 2, Figure 3 and Figure 4 and Supplementary Material Figures S5 and S6). The simulated PM_2.5 temporal trends agree with the observations from U.S. embassy sites (Figure 2, Figure 3 and Figure 4 and Figures S2–S5), especially in cases where wind speed and direction are well captured (e.g., Figure S2). In contrast, AOD time series from the AERONET site (Figures S5 and S6) show poor temporal agreement with observations, although the magnitudes are comparable. A significant underprediction of Q2 is evident at the collocated METAR sites for January; however, there is no direct evidence that this bias is affecting the performance of PM_2.5 or AOD.

The weaker agreement for AOD relative to surface PM_2.5 suggests additional uncertainties in the representation of aerosol vertical distribution and dust transport processes. Dust emissions remain one of the largest sources of uncertainty in regional air quality modeling over West Asia due to limitations in soil, vegetation, and land surface datasets used by WBDUST. These uncertainties can affect both surface PM_2.5 concentrations and aerosol column loading. While the evaluation of wind speed and direction indicates that the horizontal transport of aerosols is reasonably represented, the vertical redistribution of dust remains an additional source of uncertainty that was not directly evaluated in this study. Aerosol vertical distribution is strongly influenced by boundary layer dynamics, surface properties, radiation fluxes, and atmospheric stability, particularly over heterogeneous oceanic, desert, and urban surfaces. Consequently, uncertainties in vertical mixing and aerosol lofting may contribute to discrepancies between simulated and observed AOD, even when surface PM_2.5 concentrations are reasonably reproduced. Figure 2 and Figure 3 present the time series of PM_2.5 and AOD for January 2022, while Figure 4 and Figure 5 show the corresponding time series for June 2019. These figures provide examples of colocated aerosol observations from U.S. Embassy and AERONET sites, together with meteorological observations from nearby METAR stations (the distances between sites are indicated in the figures). Although differences in Q2 are evident, variations in the other meteorological variables are relatively small and appear to have only a limited impact on model performance in simulating aerosol concentrations and optical properties.

Figure 6 shows the spatial distribution of the NMB of PM_2.5 and AOD based on surface observations. The highest underprediction of PM_2.5 concentrations occurs at sites located outside the 10 km domain, particularly in Iraq and Saudi Arabia. However, the spatial pattern for AOD in winter shows the opposite from PM_2.5, the larger overprediction occurs within the 10 km domain and lower overprediction occurs outside it. For summer, the availability of data is more limited, making it insufficient for a similar comparison. Additionally, AOD data from AERONET are not available for all simulated cases.

Table 7. Evaluation of CAMx gaseous column abundances at a grid resolution of 20 km.

Season	Satellite	Variable	n	Obs	r	MB	NMB	RMSE	NME
Winter	OMI	NO2	34,992	3.3	0.43	0.1	3.8	3.9	51.8
		HCHO	34,930	2.6	0.00	0.6	24.9	1.3	40.8
		CO	34,967	18.7	−0.15	−1.3	−7.0	2.6	11.9
		SO2	31,291	0.1	0.19	0.3	270.7	0.9	279.5
		O3	34,992	270.2	0.87	−93.8	−34.7	107.4	36.8
	AIRS	O3	33,218	271	0.86	−91.8	−33.9	105.6	36.0
	MODIS	AOD	34,992	0.3	−0.18	−0.1	−38.8	0.1	48.3
	MAIAC	AOD	29,815	0.1	0.12	0.0	15.7	0.1	61.3
Summer	OMI	NO2	34,992	4.3	0.41	−1.2	−28.3	2.6	41.0
		HCHO	34,496	24.3	−0.21	−19.6	−80.7	84.3	87.1
		CO	34,992	18.3	0.01	−2.7	−14.7	3.1	15.4
		SO2	33,536	0.1	0.14	0.4	449.4	0.9	453.8
		O3	34,992	281.8	0.26	118.7	42.1	160.5	48.6
	AIRS	O3	33,280	284.7	0.16	116	40.8	158.7	47.9
	MODIS	AOD	34,992	0.4	0.00	−0.2	−40.8	0.3	44.0
	MAIAC	AOD	31,303	0.3	0.15	0.0	1.1	0.3	63.8

1—shaded values can be compared to literature-based thresholds; unshaded values have no such reference.

evaluates the column abundances of NO₂, SO₂, HCHO, CO, and O₃, as well as the AOD, from the CAMx 20 km simulation using satellite observations. The CAMx results show a low bias for NO₂ and CO column abundances in both seasons, indicating acceptable model performance. During summer, the model exhibits moderate underprediction of NO₂ (NMB = −14%) and CO (NMB = −28%), but both remain within the adopted performance criteria. HCHO shows a small overprediction in winter (NMB = 25%) but a substantial underprediction in summer (NMB = −80%), suggesting greater uncertainty in the representation of VOC emissions and secondary HCHO production during the warm season. Similar seasonal discrepancies have been reported in previous studies evaluating regional air quality models. SO₂ column abundances are substantially overpredicted, with NMB values ranging from 270% in winter and 450% in summer, well outside the performance criteria. However, the interpretation of this bias is complicated by known limitations on OMI SO₂ retrievals in regions with elevated CO or O₃ concentrations, where SO₂ can be underestimated or not detected [85].

O₃ columns abundances show strong spatial correlation in winter (r ~ 0.86) but considerably weaker correlations in summer (r values between 0.16 and 0.26). The model underpredicts O₃ in winter (NMB = −38%) and overpredicts it in summer (NMB = 40%), while maintaining moderate errors in both seasons (NMEs ranging from 36% to 48%). These results suggest that the model reproduces the large-scale winter distribution of O₃ reasonably well but exhibits greater uncertainty in summertime photochemical production and transport processes. The substantial summer underprediction of HCHO and moderate underprediction of NO₂ are not fully reflected in the O₃ results, which exhibit a positive bias during summer. This behavior highlights the nonlinear nature of tropospheric O₃ chemistry and suggests that factors beyond local precursor abundances, including chemical regime sensitivity, long-range transport, and boundary conditions, may contribute to the simulated O₃ columns.

AOD exhibits very low correlation (r = −0.18 to 0.15), indicating difficulties in reproducing the observed aerosol spatial patterns. Relative to MODIS, the model shows a moderate underprediction (NMB = −38% to −40%), whereas comparison with MAIAC indicates only a small bias (NMB = 1% to 15%). Despite the relatively low bias, the model exhibits moderate to large error (NME = 44–48% for MODIS and 61–63% for MAIAC), suggesting substantial uncertainty in the representation of aerosol emissions, dust transport, and aerosol option properties over West Asia.

Figure 7 shows the column abundances of NO₂, SO₂, HCHO, CO and O₃, and calculated AOD [65,86,87] from CAMx model results and satellite retrievals. The CAMx column abundances of NO₂ exhibited hotspots over urban areas and along major roads connecting these cities, reflecting significant NO₂ emissions from the transportation and industrial sectors [88]. The spatial distribution of NO₂ from OMI matches with the model results over urban areas, suggesting that the model reasonably captures the major emission sources and their spatial distribution.

For HCHO, CAMx identifies a few small hotspots over Saudi Arabia and Iran. However, OMI data appear noisy, with some hotspots detected in the same regions of Iran, though the overall patterns are less consistent. The limited agreement between CAMx and OMI may reflect uncertainties in both satellite retrievals and the representation of VOC emissions and secondary HCHO production. The CO distribution in CAMx is primarily concentrated in Saudi Arabia, with additional hotspots over urban areas in Iran. In contrast, OMI CO data depict a different spatial pattern, with low abundances over Iran and an extensive region of high abundances covering most of the domain.

CAMx SO₂ column abundances indicate several hotspots over Saudi Arabia and Iran; however, OMI did not detect the same hotspots over parts of Saudi Arabia. One important source of uncertainty is related to documented limitations in OMI SO₂ retrievals over regions with elevated concentrations of CO and O₃, where interference in the retrieval algorithm can lead to underestimation or even non-detection of SO₂. Under these conditions, model results that reasonably represent atmospheric SO₂ levels may appear artificially biased high when compared against the satellite retrievals. This also complicates the evaluation of the spatial distribution of SO₂ hotspots. In addition, SO₂ emissions over West Asia remain highly uncertain due to limitations in regional emission inventories, particularly from industrial and power generation sources. Boundary conditions may also contribute to the modeled bias. Because of the limited availability of surface SO₂ observations in the study region, it is difficult to fully isolate the dominant source of the overprediction.

For O₃, CAMx simulations produce spatial patterns similar to those observed in OMI and AIRS data, with higher column abundances at higher latitudes, aligning well with satellite observations. The northward increase in O₃ column abundances is consistent with the latitudinal gradient observed in satellite products. CAMx AOD results show numerous small hotspots over Iran, the United Arab Emirates, and several cities in Saudi Arabia. However, MODIS AOD data highlight different hotspot locations, particularly over the Red Sea and southern Oman. MAIAC AOD data show similar hotspots to MODIS but in different regions, including the Red Sea. The differences between CAMx and the satellite products likely reflect uncertainties in dust emissions and vertical transport processes, which are major contributors to aerosol loading over West Asia.

The spatial distributions of HCHO, CO, SO₂, O₃, and AOD are influenced by distinct physical and chemical processes. Elevated HCHO columns over Saudi Arabia and Iran are consistent with regions of intense anthropogenic VOC emissions associated with urban, petrochemical, and oil-production activities. Similarly, CO and SO₂ hotspots coincide with major urban and industrial centers and regions of fossil-fuel extraction and processing that have been identified as important emission sources in previous studies. The northward increase in O₃ column abundances is consistent with the large-scale latitudinal gradient observed in satellite retrievals. For AOD, the differences between CAMx and satellite products are likely related to uncertainties in dust emissions and transport, which represent a major source of aerosols over West Asia.

Overall, the meteorological evaluation indicates that temperature, wind speed, wind direction, and precipitation are reasonably represented by WRF across both seasons. Although larger uncertainties were identified for specific humidity and some cloud-radiation variables, particularly during winter, the generally good performance of the meteorological fields suggests that the major transport and dispersion processes are adequately captured (Figure 2, Figure 3, Figure 4 and Figure 5 are clear examples). Therefore, the discrepancies observed in the CAMx column abundances are likely influenced not only by meteorological uncertainties but also by factors such as emission inventories, chemical mechanisms, boundary conditions, and satellite retrieval uncertainties.

4. Conclusions

In this study, the WRF and CAMx modeling system was applied to simulate meteorological conditions and air quality over West Asia during representative winter (January 2018, 2020, and 2022) and summer (June 2017, 2019, and 2021) periods. Evaluation against available surface, satellite, and aerosol observations indicates that the modeling framework is capable of reproducing the major meteorological conditions and air-quality patterns of the region while also identifying important sources of uncertainty related to cloud processes, emissions, chemistry, and aerosol representation.

The WRF model generally reproduces the meteorological conditions over West Asia with good performance, particularly for near-surface temperature, humidity, wind speed, wind direction, and radiation variables. Temperature exhibits a slight underprediction during winter and a slight overprediction during summer, while precipitation shows moderate seasonal biases depending on the observational dataset used for evaluation. Cloud and radiation variables are generally well represented; however, cloud-related processes remain an important source of uncertainty. The poorer performance of cloud fraction, cloud water path, and cloud radiative forcing indicates limitations in the representation of cloud-radiation interactions, which contribute to the seasonal temperature biases. In winter, underestimated cloud forcing and cloud fraction are associated with excessive surface cooling, whereas in summer, uncertainties in cloud optical properties contribute to warm biases despite generally improved meteorological performance. These results highlight the importance of improving cloud microphysics and radiation parameterizations for applications in arid and semi-arid regions.

The CAMx model demonstrates generally good performance in reproducing PM_2.5 concentrations, with acceptable correlations, biases, and errors across the evaluated periods. Increasing the horizontal resolution from 20 km to 10 km reduced some biases but resulted in only modest improvements in overall model skill. The model also reproduces the major spatial patterns and hotspots of NO₂ and O₃ column abundances observed by satellite retrievals. However, larger uncertainties remain for HCHO, SO₂, CO, and aerosol optical depth, particularly during summer conditions, when biases and spatial discrepancies become more pronounced. These results suggest that uncertainties in emissions, chemical mechanisms, boundary conditions, and satellite retrieval products play an important role in the remaining model biases. Although meteorological uncertainties may contribute to some discrepancies, the generally good performance of WRF indicates that these factors are likely more important drivers of CAMx errors over West Asia.

A key finding of this study is that the season with the best meteorological performance does not correspond to the season with the best air-quality performance. While WRF exhibits its largest biases during winter, particularly for Q2 and cloud-related variables, CAMx generally performs better during this season. Conversely, larger air-quality biases are observed during summer despite the overall improved meteorological performance. This indicates that meteorological uncertainties alone cannot explain the observed CAMx errors and suggests that uncertainties in emissions, chemical processes, boundary conditions, and satellite retrievals are likely the dominant contributors to the summertime biases.

The present modeling framework remains subject to several sources of uncertainty. Important limitations include the resolution and accuracy of emission inventories, meteorological and chemical boundary conditions, and geographic datasets such as land use, vegetation, and topography. These uncertainties are particularly relevant for the representation of dust emissions, aerosol optical properties, vertical transport, and boundary-layer processes, which directly influence PM_2.5 concentrations and column-integrated quantities such as AOD. The evaluation also identified cloud-related processes, including cloud fraction, cloud water path, and cloud radiative forcing, as important sources of meteorological uncertainty. Although increasing horizontal resolution improved the representation of some local-scale features, the overall model performance remained broadly consistent across resolutions, indicating that resolution alone is insufficient to substantially reduce the remaining biases. Future improvements will likely depend on more accurate emissions, improved physical and chemical parameterizations, higher-quality boundary conditions, and the incorporation of additional observations through data assimilation and other observationally constrained approaches.

A final limitation of this study is the scarcity of observational datasets available for model evaluation over West Asia. Ground-based monitoring networks for particulate matter, trace gases, aerosol properties, and boundary-layer processes remain sparse across much of the region, limiting the ability to perform comprehensive regional-scale assessments. Consequently, part of the evaluation relies on satellite retrieval products, which introduce additional uncertainties related to retrieval assumptions, cloud contamination, surface reflectance, vertical sensitivity, and spatial representativeness. Despite these limitations, the combined use of surface observations, satellite products, and aerosol measurements provides a valuable assessment of model performance over a region where observational constraints remain limited. Future improvements in both model evaluation and air-quality prediction will require expanded monitoring networks, additional vertical profile observations, and more comprehensive long-term measurements of trace gases and aerosols across West Asia.