Vertical Ozone Concentration Profiles in the Arabian Gulf Region during Summer and Winter : Sensitivity of WRF-Chem to Planetary Boundary Layer Schemes

Air quality in the Middle East is changing due to extensive land conversion, intense industrialization and rapid urbanization. In this study, we analyze data from an ozonesonde station operated in Doha, Qatar, by the Qatar Environment and Energy Research Institute (QEERI). Ozonesondes were launched weekly at 13:00 LT (10:00 UTC) during a summer month (August 2015) representative of extremely hot and humid atmospheric conditions and during a winter period (January–February 2016) representative of cool conditions in the area. Unlike similar studies in the region, this work focuses on the lower troposphere and combines high frequency vertical measurement data with the use of the Weather Research Forecasting model coupled with Chemistry (WRF-Chem). A sensitivity study was conducted to identify the most representative planetary boundary layer (PBL) parameterization. Although all three parameterizations that were examined produced similar results, the Yonsei University (YSU) PBL scheme was found to be statistically superior. Comparisons of model predictions against observations show high correlation coefficients and encouragingly low biases in all meteorological variables. During wintertime, ozone is well predicted overall (fractional bias = –0.1). Results from the summertime comparison are more challenging and point towards possible biases in the anthropogenic emission inventory of the Middle East, especially for rapidly-changing urban environments.


INTRODUCTION
Surface ozone-one of the main criteria pollutants set forth by the US EPA (Environmental Protection Agency)is formed photochemically through oxidation of VOCs (volatile organic compounds) in the presence of nitrogen oxides (NO x ) (Jacob, 1999).Tropospheric ozone is of major concern due to its negative impact on human health and global warming.The influence of the free troposphere on planetary boundary layer (PBL) ozone has been reported in several studies (Trickl et al., 2003;Cui et al., 2011).In the Middle East, apart from dust-rich atmospheric conditions (Saraga et al., 2017), high levels of surface ozone formation are also often observed due to intense sunlight and a considerable amount of (mainly anthropogenic) precursors in urban environments.Although the Middle Eastern region has been characterized as a tropospheric ozone "hotspot" (Lelieveld et al., 2009), very few measurement and modelling studies in the region exist to date (Li et al., 2001;Lelieveld et al., 2009;Zanis et al., 2014;Spohn and Rappenglück, 2015).Li et al. (2001) used the GEOS-Chem global atmospheric chemistry model and showed unusually high levels of ozone in the Middle East (in the middle to upper troposphere) resulting mainly from large-scale subsidence inversion and production in the upper troposphere.Lelieveld et al. (2009) using the EMAC-MESSy global model, calculated an average of 80 ppbv of summertime ozone concentration in the middle to upper troposphere (from 6 to 10-12 km).They attributed this to long-range transport of pollution, a stratospheric-tropospheric exchange stronger than expected, and a significant amount of NO x production from lightning.Similarly, Zanis et al. (2014) used the same model (EMAC) along with satellite data for a 12-year period indicating a pool of ozone in the middle to upper troposphere above the Eastern Mediterranean while highlighting a significant contribution from stratospheric ozone.Spohn and Rappenglück (2015) used in situ measurements (ozonesondes) in the Middle East region to explore potential sources of ozone in the upper troposphere.In agreement with the above modelling studies, they observed an average of 80 ± 13 ppbv in the 6-12 km altitude range and a maximum of ~100 ppbv of ozone at around 8 km in height during August.
To better understand episodic air pollution, studies have widely made use of numerical weather prediction models and air quality/chemical transport models.Complex geographical regions require accurate representation of both the meteorology and the physical-chemical processes involved.In some model applications, the grid resolution is relatively low, or the meteorological and chemical processes are offline connected.Such approaches have the disadvantage of not accounting accurately enough for the sub-grid scale phenomena or the interactions and feedback between meteorology and chemistry which could be a significant loss of information especially in polluted urban environments with complex meteorological regimes.Furthermore, planetary boundary layer (PBL) processes play a crucial role in simulations of the lower atmosphere.Atmospheric turbulence can affect vertical mixing and the distribution of chemical species concentrations.The PBL characteristics can vary a lot, and thus, different parameterizations exist, which could represent the state of the atmosphere in various ways affecting the predicted meteorology and chemistry of the large-scale model.
In this work-unlike previous studies in the region-we focus on the boundary layer and lower troposphere (0-6 km) and combine in situ data analysis of ozonesondes with a 3-D regional online-coupled meteorology-chemistry model.We explore two meteorologically different periods, a summer (August 2015) and a winter (January/February 2016) time period, and investigate the sensitivity of the model predictions to different PBL parameterizations in simulating meteorological and air quality parameters in the coastal urban environment of Doha, Qatar, using high spatial resolution over the state of Qatar.

In Situ Measurements
During the period of study, ozonesondes were launched from the northeast of Doha (25.361°N, 51.481°E) once per week, at 1 p.m. local time (LT).Data quality assessment was performed on all ozonesondes launched during the study period, following the standardized data validation methodology of the US EPA (US Environmental Protection Agency, 2000).Each balloon setup was equipped with a Global Positioning System (GPS) reporting coordinates (latitude, longitude) every second along with an iMet-1 radiosonde (Wierenga and Parini, 2005), which measured ambient pressure, temperature, wind direction, wind speed (derived from GPS coordinates), relative humidity (RH), and altitude.An electrochemical concentration cell (ECC) ozonesonde was attached to the weather balloon, providing ozone data with an accuracy ranging between 5% and 12%, depending on the altitude (EN-SCI;Environmental Science, 2016).To complement the ozonesonde data, the vertical stratification and mixing heights were also monitored with lidar-ceilometer observations.Atmospheric backscatter measurements were carried out with a Vaisala CL51 ceilometer on QEERI's premises in Education City, Doha (25.329°N, 51.426°E) (Bachour and Perez-Astudillo, 2014a, b).The instrument operated uninterruptedly (except for some occasional missing data), which allows for a continuous determination of PBL heights and their evolution.The CL51 ceilometer works with 910 nm nearinfrared laser pulses emitted at a frequency of 6.5 kHz, reporting backscatter profiles every 36 seconds with a vertical resolution of 10 m and maximum range of 15 km.Its operation is based on the elastic lidar technology: Laser pulses are emitted, and, by measuring the time taken for the reflected light to travel back to the instrument, the distance to the reflecting element is calculated; elastic lidars only detect reflected photons with the same energy (wavelength) as the emitted light.In ceilometers, the pulses are sent to the atmosphere (vertically, as in the case of this study, or at an angle), where suspended particles absorb or scatter the light.By collecting the light reflected along the same line of emission and reading out the backscattered signal intensities at predetermined time windows, the concentrations of atmospheric contents at different distances (heights) along with backscatter profiles can be derived from the collected signal intensities.The most widely used identification algorithm consists of calculating the gradient of the range-corrected ceilometer backscatter signal with height.Excluding clouds, which give the largest signals, large changes in particle concentration indicate different layer boundaries.The PBL height or boundary is usually identified by the sharpest gradient, followed by a relatively particle-free atmosphere above, and with the aid of the knowledge of the expected diurnal evolution of the PBL.

Model Description and Simulation Experiments The WRF-Chem Model Setup
Generating accurate predictions of the spatial and temporal variation in species concentrations in a system as complex as the arid urban environment of Doha, Qatar, which involves numerous physical and chemical processes occurring simultaneously, requires the use of a numerical chemical transport model.Within this work, the three dimensional meteorology-chemistry model WRF-Chem version 3.7 (Weather Research Forecasting with Chemistry (Grell et al., 2005;Fast et al., 2006)) was employed over the entire Arabian Peninsula region with an enhanced grid resolution over the state of Qatar (Fountoukis et al., 2016).The WRF-Chem model simulated the three basic components: the processing of emissions of atmospheric constituents (gases and aerosol particles); transport; and the physicochemical transformations of atmospheric species.The model was applied over the Middle Eastern area in a domain of 3-D grids on a two-way nesting configuration in which three domains at different grid resolutions communicate with each other and are run simultaneously (Fig. 1).Information concerning species concentrations propagates into and out of all computational domains during the model integration.The parent domain used a 50 km × 50 km grid resolution, the intermediate nested domain (focusing on the Arabian Desert) used a 10 km × 10 km resolution, and the third domain was configured over the region of Qatar and was resolved at 2 km × 2 km.This grid nesting capability of WRF-Chem allows for a computationally efficient model capable of spanning large areas in which regional transport of pollutants is important while providing fine resolution in select areas to address small-scale features.The altitude coordinate was discretized into 28 vertical layers in all three computational domains, extending from the surface to approximately 100 hPa.The model simulates two periods of 30 days each, representative of a typical summer (1-30 August 2015) and winter (19 January-17 February 2016) period in Qatar.Each simulation run was initialized and permitted to simulate 10 days in advance of the model period; this span of time is considered to be spin-up time and was excluded from the analysis to limit any extraneous negative effects due to the initial conditions.Boundary conditions for the outermost domain are taken from idealized profiles specified in the chemistry routines.WRF-Chem was set to perform all simulation runs on a Lambert map projection.Dynamic meteorological and static geographical data were generated by the Global Forecast System (GFS) and used to drive WRF.

Emission Inventories
The anthropogenic emissions data set that is used in all three domains comes from the Emission Database for Global Atmospheric Research (EDGAR) (http://www.mnp.nl/edgar/introduction) and the REanalysis of the TROpospheric (RETRO) chemical composition (http://retro.enes.org).Both EDGAR and RETRO provide global annual emissions for several precursor and greenhouse gases on a 0.1° × 0.1° (EDGAR) or a 0.5° × 0.5° (RETRO) resolution.The latest version of EDGAR emissions is used, namely the HTAP_V2 (Hemispheric Transport of Air Pollution emissions -version 2; http://www.htap.org/)anthropogenic emissions, which include non-methane volatile organic compounds (NMVOCs), nitrogen oxides (NO x ), ammonia (NH 3 ), carbon monoxide (CO), sulfur dioxide (SO 2 ), black carbon (BC), and organic carbon (OC).These emissions are provided for several source sectors: transport, energy, industry, agriculture, and the residential sector.Table 1 shows a summary of the HTAP anthropogenic emissions for each source sector.Biogenic emissions are calculated online during runtime through the Guenther et al. (1994) approach, in which the USGS land use classification is used to generate emissions with the WRF pre-processing system (WPS).Dust was simulated using the US Air Force Weather Agency (AFWA) emission scheme, which incorporates the MB95 dust emission parameterization (Marticorena and Bergametti, 1995) with typical airborne dust size distributions (Kok, 2011) and calculates dust fluxes via: where E is the erodibility factor (Ginoux et al., 2001), α is the sandblasting efficiency, c is an empirical proportionality constant, g is the gravitational acceleration, ρ α is the air density, u * is the friction velocity, and u *t is the threshold friction velocity.

Simulation Experiments
The GOCART (Georgia Institute of Technology-Goddard Global Ozone Chemistry Aerosol Radiation and Transport) aerosol scheme (Ginoux et al., 2001;Kok et al., 2011) was used in all simulations along with the RACM (Regional Atmospheric Chemistry Mechanism) chemistry scheme (Stockwell et al., 1997;Geiger et al., 2003).
The physics part of WRF-Chem stems from the WRF model, a nonhydrostatic meteorological mesoscale model that includes parameterizations of land surface, planetary boundary layer, and cloud processes.The base case simulations of this work include the Grell 3D cumulus parameterization (Grell and Devenyi, 2002), the Lin microphysics scheme (Chen and Sun, 2002), the 5-layer thermal diffusion Land Surface Model, the Yonsei University boundary layer scheme (Hong et al., 2006), the Goddard shortwave radiation scheme (Chou et al., 1998), the revised MM5 Monin-Obukhov surface layer scheme, and the Rapid Radiative Transfer Model longwave radiation scheme (Mlawer et al., 1997).
In general, two closure scheme approaches exist for the determination of turbulent fluxes from atmospheric variables in a grid cell of a numerical weather prediction model, namely, local and non-local closure parameterizations (Stull, 1988).The local closure scheme, assuming turbulence resembles molecular diffusion, estimates mean grid cell atmospheric parameters and their gradients.Non-local closure schemes parameterize vertical transport through advection among model grid cells at adjacent levels.Relatively few studies (Yeramilli et al., 2010;Cuchiara et al., 2014) exist on the performance of WRF-Chem with respect to various available PBL schemes and none in the Middle East to the best of our knowledge.In this work, we compare the YSU parameterization (Hong et al., 2006), a non-local PBL scheme that has been shown to work well especially in coastal urban areas (Yeramilli et al., 2010;Cuchiara et al., 2014), with the new local Mellor-Yamada-Nakanishi-Niino level 2.5 (MYNN) (Nakanishi and Niino, 2009) scheme and the Grenier-Bretherton-McCaa (GBM) PBL, which is a turbulent kinetic energy (TKE) scheme that was recently added in WRF and has shown to perform well (Perlin et al., 2014).A number of other PBL schemes  have been tested to date in urban coastal areas without showing any clear superiority to the YSU scheme (Yeramilli et al., 2010;Cuchiara et al., 2014) and thus are neglected in this comparison.

MODEL EVALUATION
The prediction skill of WRF-Chem is quantified in terms of the mean absolute gross error (MAGE), the mean bias (MB), the fractional error (FERROR), the fractional bias (FBIAS), and the correlation coefficient (R): where P i is the model predicted value for data point i, O i is the observed value and n is the total number of data points.We consider weekly midday launches, when ozone is expected to peak for a typical weekday, with a weekly frequency.Tables 2 and 3 summarize the prediction skill metrics of WRF-Chem against the balloon measurement data of meteorological parameters and ozone concentrations for the summer and winter period, respectively.Fig. 2 shows the planetary boundary layer height time series while Figs. 3 and 4 show the comparison of the simulated vertical profiles of air temperature and relative humidity against those measured during both periods of study.
The PBL height estimated by the three sensitivity runs shows the different mixing in the lower atmosphere that the model predicts for each scheme (Fig. 2), which is expected to have a vital impact on air quality predictions in this altitude range.Fig. 2 also includes the measured backscatter profiles indicating observed boundary layer heights by identifying large changes/gradients in particle load.The detection of clouds, aerosols, and atmospheric boundary layers is based on changes in the measured concentrations with height and time, using the backscatter profile measured by the lidar (Kovalev and Eichinger, 2004).During nighttime, the simulated PBL height estimated by YSU ranges between 30 and 300 m in summer and between 40 and 500 m in winter.The MYNN nocturnal PBL height is somewhat higher (70-500 m in summer and 90-600 m in winter), while the GBM is the scheme that calculated the lowest PBL on average.The height at which the balloon flight measured a certain distinct change in temperature and relative humidity is shown in Fig. 2 by a circle at the respective time of flight.This further assists in identifying the correct measured PBL height profile from the ceilometer measurements.During daytime, the differences between the three sensitivity runs are larger as a consequence of the strong convective turbulence.The largest differences are seen on 27 January with the YSU scheme calculating a peak PBL height of 1450 m (compared to 850 m from MYNN and 740 m from GBM) and on 20 January when the MYNN peak is estimated to be three hours later than the YSU peak prediction.These differences, caused by inherent differences in the different parameterizations, are expected to have induced some changes in the predicted ozone concentrations August and 10 February), it is impossible to draw any definite conclusion for all days based on the ceilometer data.
In general, the air temperature is in good agreement with the ambient data, with a correlation coefficient of 0.99 for all three simulation experiments.Some small differences among the predictions of the three sensitivity runs are seen in the 0-2 km altitude range driven by the different vertical transport calculations within the PBL parameterizations.As the balloons were launched during midday and the PBL exhibits a distinct diurnal variation, different PBL schemes estimate more pronounced differences in air temperature in this altitude range and during this time of day.For example, on 27 January 2016, when the biggest differences among the three sensitivity runs were seen, the YSU simulation predicts a mean temperature of 15°C below 2 km altitude, while the MYNN and GBM simulations predict 12.2°C and 15.9°C, respectively, compared to an observed value of 14.4°C.The largest model-measurement discrepancy for all three model runs is seen on 5 August 2015, when all three simulations underestimate the mean temperature below 2 km by approximately 2.5°C.This influences the relative humidity predictions during that day, as an underestimation of temperature may be closely linked to an overestimation of RH.
RH profiles simulated by the three model experiments show larger differences than the temperature profiles, even at altitudes above the boundary layer.During summertime, the MYNN PBL consistently underestimates RH, showing the largest mean bias (-12%), while the YSU scheme predictions have the lowest fractional bias (-0.06).This is in agreement with previous studies (Yerramilli et al., 2010) that showed that the YSU scheme predicts more realistic RH profiles for both stable and unstable conditions.On average, all three PBL runs overestimate the RH during wintertime, with the MYNN PBL run showing the largest error (mean error = 13.4%).These differences in RH are most likely due to differences in the wind direction and wind speed simulated by the three model experiments at various altitudes.It should be noted that the temperature, humidity, and wind speed distributions in the lower atmosphere are also influenced by the land-surface model and the surface layer scheme adopted within each simulation.For consistency and in order to test the sensitivity of our model predictions to the PBL scheme only, we have used the same land-surface model in all simulation experiments, namely, the thermal diffusion scheme.Furthermore, the YSU and GBM PBL schemes are coupled with the revised MM5 Monin-Obukhov surface layer scheme, while the MYNN PBL is coupled with the MYNN surface layer option within WRF-Chem.During all summertime balloon flights, there is a marked change (increase) in ambient relative humidity above the boundary layer.On 12, 19, and 26 August, both the GBM and YSU runs capture the RH profile changes relatively well.The MYNN simulation fails to reproduce this distinct change, especially during the 19 August flight, and the GBM also fails during the 5 August balloon flight.On 12 August, the observed RH changes sharply above 1.5 km, but all three model experiments fail to reproduce this behavior as the altitude of the balloon increases.This is due to subsidence inversion as well as transport of air masses from the Indian monsoon, which are highly humid and have been well documented before in the area for the same period (Spohn and Rappengluck, 2015).All model runs underestimate this sharp boundary layer change at around 1.5 km altitude (also evident in Fig. 2).During winter, all three simulations well predict the abrupt decrease in RH above 3 km (on 27 January), which is also seen during some flights in August above 5 km and is attributed to air masses (large-scale flows) transported from a less humid environment (e.g., the Eastern Mediterranean) and a large-scale subsidence inversion during summer that has been reported before (Kourtidis et al., 2002;Lelieveld et al., 2002Lelieveld et al., , 2009;;Spohn and Rappenglück, 2015).Fig. 5 shows the predicted wind speed and wind direction over the Qatari region at 1 p.m. LT on 27 January as calculated by the YSU simulation at the surface and at an altitude of 3 km.Clearly, there is a change in the wind direction and speed between the surface and aloft.The model predicts strong winds originating from the southwest of Doha at an altitude of 3 km, while at the surface, the predominant winds are from the north.
The water vapor mixing ratio measurements show a distinct pattern within the vertical profile with a sharp decrease up to approximately 2 km in most of the balloon flights and a somewhat constant value at altitudes between in 2 and 6 km (Fig. 6).All simulations capture this concentration pattern qualitatively with the exception of the MYNN simulation during 19 August.On 27 January, however, the vertical profile of the observed water vapor mixing ratio is notably different than in the rest of the flights, especially in the altitude range of 1-2 km, showing a sharp increase in the mixing ratio with altitude, which is only reproduced by the GBM parameterization.Interestingly, all three model runs underestimate the mean observed water vapor mixing ratio during summer (by 1.7 g kg -3 for YSU, 2.8 g kg -3 for   GBM, and 3.5 g kg -3 for MYNN) and overestimate it during winter (by approximately 1 g kg -3 across schemes).
Fig. 7 shows a comparison of observed ozone concentration profiles with WRF-Chem predictions.In general, the formation of ozone is affected by the local weather conditions and the horizontal and vertical transport processes.The model calculated the wintertime ozone concentrations reasonably well while showing larger discrepancies with balloon measurements during the summer.During winter, the YSU scheme performs on average statistically better than the MYNN or GBM, with a fractional bias of -0.13 and a mean error of 7.4 ppb; with the exception of 20 January, all three sensitivity runs produce very similar wintertime ozone concentrations.This is in agreement with Yeramilli et al. (2010) and Cuchiara et al. (2014), who found the YSU scheme superior in predicting vertical ozone concentrations in urban coastal environments.Within the boundary layer, the model underestimates ozone concentrations by an average of 1-10 ppbv.During the summer period, the model underestimation is higher.All three sensitivity runs underpredict ozone on average by 35%.A similar underprediction (30%) is also reported in Cuchiara et al. (2014) during the month of October in the altitude range of 0-5 km.As August is the hottest month of the year in Qatar, warm conditions lead to strong reaction rates and intense ozone production.All model runs failed to predict the sharp increase in O 3 with altitude in the 0-1 km range that was observed during both 5 and 19 August due to an inversion layer and suppressed mixing.The underestimation of temperature by the model on 5 August could partly explain the underprediction of ozone formation by all three schemes.However, the overall model performance during August suggests that the main reason for the model-measurement discrepancies with regard to ozone is not related to the meteorology or the choice of PBL scheme.Fig. 8 shows the mean spatial distribution of surface ozone concentration predicted at 1 p.m. LT during August with the different PBL schemes.The patterns of ozone distribution are similar among the three simulations.Overall, all three simulations predict an average of 45-55 ppbv throughout most of Qatar.The MYNN scheme predictions are somewhat higher for the south of the domain compared to the YSU and GBM results, likely originating from the different surface layer option employed by the MYNN simulation.Both YSU and MYNN predict a peak ozone concentration at the west-northwest of 55-60 ppbv, while the GBM run predicts that to a lesser extent.These differences are probably related to variations in the predicted wind speed and wind direction patterns as well as the temperature and other meteorological parameters estimated by the three schemes.The peak ozone predicted at the west-northwest of Qatar is due to long-range transport over population centers and oil and gas development regions in the northwest as well as a strong convergence in the surface level flow within 10-20 km from the western shoreline.Fig. 9 shows this north-south line of strong convergence predicted throughout the western side of the country.The illustration of the hourly change in ozone concentrations during August over Qatar shows that a large part of the average concentrations of ozone at 1 p.m. LT predicted in the metropolitan area of Doha has actually been transported from outside the domain (often more than 60 ppb, mostly from the north, in line with the observed wind profiles).Long-range transport is responsible for most of the ozone ground levels in Doha, with the locally produced ozone being most likely underestimated by the model.
As a secondary pollutant, ozone is formed by reactions involving solar radiation and precursor emissions of VOCs and nitrogen oxides.Sensitivity simulations for both seasons (not shown here) indicate that the impact of biogenic    emissions on vertical ozone profiles, as calculated online by the Guenther scheme, is low (ranging between 0.2% and 4% in summer and 0.3% and 9% in winter) due to limited vegetation in the area.Anthropogenic emissions, on the other hand, have continuously changed during recent years due to a significant population increase in Qatar as well as in other large urban environments of the Middle East.Industry is the largest emission source for CO, anthropogenic PM, SO 2 , NMVOC, and NH 3 (Table 1).Residential activities have a small contribution in all species, ranging from 0.1% to 10%.Ammonia emissions are mainly from industry (53%) in Qatar, as agricultural activities are limited in this arid environment.Energy (power industry) is the largest source (40%) of NO x emissions, followed by industry (31%) and transportation (29%).As these emissions are for the year 2010 and given that the population of Doha has risen considerably (~50%) since then, it is expected that the current contribution of transportation is higher.In fact, a comparison of observed surface NO x concentrations at the QEERI ground monitoring station in Doha with the model shows an average underprediction of NO x levels by a factor of 3 during summertime.

CONCLUSIONS
For the first time, we analyzed vertical profiles of ozone concentration in the lower troposphere of the Middle East by combining a regional chemical transport model (WRF-Chem) with balloon measurements in the coastal urban environment of Doha, Qatar.We tested the sensitivity of the model to 3 different boundary layer schemes during both a summer and a winter period.A triple nested model configuration with high spatial resolution was selected over the domain of interest.High-frequency monitoring of the vertical profiles of temperature, relative humidity, water vapor mixing ratio, and ozone concentration was performed via balloon launches, and the results were compared with the respective model predictions.
Overall, all three boundary layer schemes produced similar results for the studied meteorological parameters during both seasonal periods.During winter, the YSU simulation was statistically superior to that of MYNN or GBM in predicting vertical ozone profiles, with a fractional bias of -0.13 and a mean error of 7.4 ppb.During summertime, there was a tendency to underpredict ozone (~35%) by all three sensitivity runs during most of the flights.The predicted patterns of the ozone spatial distribution at the surface were similar among the three simulations, which estimated an average of 45-55 ppbv throughout most of Qatar.Model results suggest that the YSU scheme is the right boundary layer option within WRF-Chem for the studied region, as it improves model predictions, especially during wintertime.However, the discrepancies between the model predictions and the observations are caused not by a particular boundary layer parameterization but by other issues (e.g., emissions or chemistry).Future work should focus on a careful examination of the emission inventory for the Middle East and most likely a refinement of the relevant emissions, mainly anthropogenic ones, especially in fast-changing urban environments similar to Doha.

Fig. 2 .
Fig. 2. Comparison of planetary boundary layer height time series simulated within WRF-Chem against observations (ceilometer and radiosonde data) during summer and winter.

Fig. 3 .
Fig. 3. Comparison of simulated vs. observed (red dots) air temperature vertical profiles with different PBL schemes used within WRF-Chem (YSU: black line, MYNN: blue line, GBM: orange line) during summer and winter.

Fig. 4 .
Fig. 4. Comparison of simulated vs. observed (red dots) ambient relative humidity vertical profiles with different PBL schemes used within WRF-Chem (YSU: black line, MYNN: blue line, GBM: orange line) during summer and winter.

Fig. 5 .
Fig. 5. Spatial distribution of wind speed (m s -1 ) and wind direction predicted by WRF-Chem (YSU scheme) at 1 p.m. on 27 January 2016 at the ground layer (left panel) and at 3 km altitude (right panel).

Fig. 6 .
Fig. 6.Comparison of simulated vs. observed (red dots) water vapor mixing ratio vertical profiles with different PBL schemes used within WRF-Chem (YSU: black line, MYNN: blue line, GBM: orange line) during summer and winter.

Fig. 7 .
Fig. 7. Comparison of simulated vs. observed (red dots) ozone vertical profiles with different PBL schemes used within WRF-Chem (YSU: black line, MYNN: blue line, GBM: orange line) during summer and winter.

Fig. 8 .
Fig. 8. Average spatial distribution of ozone concentration at the ground layer predicted by WRF-Chem at 1 p.m. during August using three PBL parameterizations (panel a represents the YSU results, b the MYNN and c the GBM simulation).

Fig. 9 .
Fig. 9. Spatial distribution of wind speed (m s -1 ) and wind direction predicted by WRF-Chem (YSU scheme) at 1 p.m. on 4 August 2015 at the ground layer.

Table 1 .
Summary of hemispheric transport of air pollution (HTAP) emissions (in tons month -1

Table 2 .
Comparison of WRF-Chem with balloon measurements in the Arabian troposphere during summer.

Table 3 .
Comparison of WRF-Chem with balloon measurements in the Arabian troposphere during winter.
as well as meteorology predictions.Although there seems to be some superiority of the YSU scheme in successfully representing the PBL height during certain days (e.g., 19