The KAUST campus is located in the western Arabian Peninsula, on the eastern coast of the Red Sea (22.3∘ N, 39.1∘ E). This area is affected by local dust storms originating from surrounding inland deserts, by distantly generated dust arriving from northeastern Africa through the Tokar Gap (see, for example, Kalenderski and Stenchikov, 2016; Albugami et al., 2019; Kumar et al., 2019), and by dust from as far away as the Tigris and Euphrates regions (Parajuli et al., 2019). Therefore, dust is present in the atmosphere over the study site for the entire year.

Although our focus in this study is on dust aerosols, which are the dominant aerosol over the study site (Jish Prakash et al., 2015; Parajuli et al., 2019; Ukhov et al., 2020b), some additional aerosol types also contribute to the aerosol loading at KAUST. Our site is located on the coast; thus, sea salt aerosol, which is of natural origin, inevitably contributes considerably to the atmospheric aerosol loading. Furthermore, the study site has several industrial areas nearby that produce anthropogenic emissions of sulfur dioxide (SO2) and black and organic carbon (BC and OC) (Ukhov et al., 2020b).

Because the site is located exactly at the land–sea boundary, some unique small-scale processes exist that affect the local climate of this region. For instance, land and sea breezes affect the distribution of dust in the atmosphere over the study site. The desert land heats up during the day, which consequently heats the surface air above the land. This warm air mass rises due to convection, creating a local pressure “low” at the surface. The cooler and more moist air over the Red Sea then flows towards the low-pressure zone, thus forming sea breezes (Simpson, 1994; Miller et al., 2003; Davis et al., 2019). During the night, this flow is reversed to form land breezes, when the land surface temperature cools more quickly than the sea surface temperature. Because these breezes are driven by the thermal contrast between the land and the sea, their strengths vary by season. These breezes are further enhanced because of their coupling with slope winds generated on the Sarawat Mountains, which run along the western coast of the entire Arabian Peninsula (Davis et al., 2019).

Land and sea breezes affect dust aerosol emissions and transport in our study region. When the land and sea breezes are strong, they can cause dust emission from the coastal regions. Although breezes are not responsible for long-range transport, they can affect the local distribution of dust aerosols over the study site.

Because dust/aerosols are present over the study site for most of the year, they can also interact with the meteorology and thus affect atmospheric winds and temperature at different timescales (Jacobson and Kaufman, 2006; Rémy et al., 2015). Land or sea breezes are strongly coupled with dust/aerosols and temperature variability, especially near the surface (Crouvi et al., 2017).

We use several datasets to evaluate our model simulations and derive the average season profiles of aerosol loading and surface winds during 2015–2016, as described below.

We use WRF-Chem (v3.8.1) with some recent updates (Ukhov et al., 2020a) for simulating the emission and transport of dust and other aerosols at high resolution at the study site. The innermost domain (d03), which is marked by a red box in Fig. 1, is centered at KAUST and has a fine resolution of 1.33 km, which is required to resolve the essential features of local wind circulation and breezes. The innermost domain is encompassed by a second domain (d02) with a resolution of 4 km that covers the entire Arabian Peninsula. Although the western boundary of domain d03 appears close to that of d02, there are 40 grid cells in between, which is 10 times higher than generally recommended and is sufficient to ensure a smooth transition across the boundaries. While a further westward extension of d02 could be desirable to better resolve the synoptic weather phenomena across the Red Sea, e.g., through the Tokar Gap (Kalenderski and Stenchikov, 2016), such phenomena have a minor impact on the diurnal-scale local sea breeze circulation in our site, which is the focus of our study. To allow full aerosol exchange and cover all major sources of dust in the region, we nest the two inner domains within a larger domain (d01) with a 12 km resolution, which covers the entire Middle East and northern Africa (MENA) region shown in Fig. 1. The key physics and chemistry options used in WRF-Chem are presented in Table 1.

The model top is set at 100 hPa and has 30 vertical levels between ∼20 m and 16 km. To represent winds better, we apply “grid nudging” on the u (zonal velocity) and v (meridional velocity) components of wind above the planetary boundary layer (PBL) in all three domains (Parajuli et al., 2019). We do not use any convective parameterization scheme and resolve deep convection in the innermost domain. We employ two-way nesting, which means that the parent domain provides boundary conditions for the nest, and the nest provides feedback to the parent domain. The model time steps are set to 72, 24, and 8 s for the three domains d01, d02, and d03, respectively.

Several studies compare the performance of PBL schemes in WRF, showing mixed results under different model settings (e.g., Saide et al., 2011; Fountoukis et al., 2018; Fekih and Mohamed, 2019). However, these studies have not directly compared the aerosol vertical profiles. Preliminary results showed that the choice of the PBL parameterization did not have a significant impact on the vertical distribution of aerosols in our case. In our simulations, we use the YSU PBL scheme, which is one of the most commonly used schemes, as suggested in the literature (e.g., Fountoukis et al., 2018; Fekih and Mohamed 2019).

We use high-resolution operational analysis data from ECMWF (∼15 km) to provide initial and boundary conditions in our model, which are updated every 6 h. The sea surface temperature (SST) values are also updated in our simulations, using the same ECMWF dataset.

We employ the Global Ozone Chemistry Aerosol Radiation and Transport (GOCART) aerosol scheme in our simulations (Chin et al., 2002). To calculate dust emissions, we use the AFWA dust scheme, which follows the original GOCART dust scheme (Ginoux et al., 2001) modified to account for saltation (Marticorena and Bergametti, 1995; LeGrand et al., 2019). It is important to represent the dust sources at a fine scale to capture the smaller-scale physical processes accurately. Therefore, we use a recently developed high-resolution sediment supply map (SSM) as the source function (Parajuli and Zender, 2017; Parajuli et al., 2019) in all three model domains. We adopt the tuning process of the dust model described in Parajuli et al. (2019). We tuned the model against CALIOP DOD, and the same tuning coefficients obtained from Parajuli et al. (2019) are used in all domains, including the added third domain, and are 0.136, 0.196, 0.120, and 0.110 for DJF, MAM, JJA, and SON, respectively.

We consider dust, sea salt, sulfate, and black and organic carbon (BC and OC) aerosols in our simulations. Biomass burning and biogenic aerosols are not important over the region, and thus we do not include them.

Sea salt emissions in WRF-Chem follow the parameterization developed by Monahan et al. (1986) and Gong (2003). In this parameterization, the rate of sea salt emissions produced via whitecaps and wave disruption is given as a function of particle size and 10 m wind speed.

We take the anthropogenic emissions of OC and BC from the most recent version of EDGAR (Emission Database for Global Atmospheric Research) database v4.3.2 available at 0.1∘ × 0.1∘ resolution (Crippa et al., 2018). The EDGAR database is a global database that provides gridded emission maps of several greenhouse gases and air pollutants from 1970 to 2012. We use OC and BC emissions data from 2012.

Sulfur dioxide (SO2) is of particular concern because it chemically transforms in the atmosphere into secondary sulfate, which is an important and influential aerosol at our study site (Ukhov et al., 2020b, c). To achieve a more accurate representation of sulfate aerosols, we use the SO2 emissions from a time-varying (monthly) inventory developed by NASA for the same year (2015). This SO2 inventory is developed by combining satellite-based estimates from the Ozone Monitoring Instrument (OMI) with the ground-based inventory developed by the Task Force on Hemispheric Transport Air Pollution (TF HTAP) (Janssens-Maenhout et al., 2015), which provides a more accurate gridded emission dataset with greater spatial and temporal coverage. The data have global coverage with 0.1×0.1∘ resolution (Liu et al., 2018). This dataset does not account for SO2 emissions produced by ships; therefore, we take ship SO2 emissions from the EDGAR v4.3.2 dataset. OMI-HTAP emissions in WRF-Chem are satisfactorily reproduced by the observed SO2 loading in the Middle East region (Ukhov et al., 2020c).

We activate both gas and aerosol chemistry in our simulations (gaschem_onoff = 1, aerchem_onoff = 1) and apply the aerosol chemistry options in all three domains.

To determine the contribution of each aerosol species to total AOD, we modify the WRF-Chem code, mainly the Fortran subroutines in optical_driver.F and chem_driver.F located in the chem folder. For this purpose, we calculate aerosol optical properties twice, first with the mixture containing all aerosols and second after removing a specific aerosol. This calculation is implemented in the subroutine “optical_averaging”. Thus, we obtain the contribution of specific aerosol species to total AOD by subtracting the AOD obtained without a specific aerosol from the total AOD calculated when all aerosols are accounted for.

We calculate the total aerosol concentration (TAC) in µgm-3 by summing up the individual concentrations of all aerosol species. The equation used to calculate the total aerosol concentration from the standard output variables of WRF-Chem is presented below. TAC (µgm-3)=[(DUST_1+DUST_2+DUST_3+DUST_4+DUST_5)+(SEAS_1+SEAS_2+SEAS_3+SEAS_4)+(OC1+OC2)+(BC1+BC2)+P10+P25]×1/ALT+sulf×1/ALT×1000×96/29, where DUST_1…DUST_5 are the dust mass mixing ratios (µgkg-1) in five different size bins; SEAS_1…SEAS_4 are the sea salt mass mixing ratios (µgkg-1) in four different size bins; P10 and P25 are other anthropogenic PM10 and PM2.5 mass mixing ratios (µgkg-1), respectively; OC1 and BC1 are mass mixing ratios (µgkg-1) of hydrophobic organic carbon and black carbon, respectively; OC2 and BC2 are mass mixing ratios (µgkg-1) of hydrophilic organic carbon and black carbon, respectively; sulf is the SO4 volume mixing ratio (ppmv), ALT is the inverse of air density (m3kg-1), and 96/29 is the ratio of the molecular weights (gmol-1) of sulfate and air.

We conduct the model simulations for the entire year of 2015 on a monthly basis (for computational reasons). For each month, the model simulations start a week before the month begins, and we discard the data from this week as spin-up. We use data for 2015 only for the comparison of the model results with other datasets. However, we use the entire 2 years of data (2015–2016) to derive the seasonal profiles. Because we aim to explore the diurnal cycles, we use hourly model output data for analysis. While comparing point measurements (lidar and meteorology data) with gridded datasets, we use data from one grid cell containing the KAUST site from all gridded datasets.

Figure 2a shows the mean diurnal cycle (2015/16) of station-measured surface wind speed at the study site. The surface winds reach a peak around noon UTC (15:00 local time) for all seasons except winter, consistent with the results of Davis et al. (2019). The aforementioned sea breezes cause these wind peaks in the afternoon. Note that these sea breezes originate at sea and advance landward to reach the coast only later in the afternoon (Estoque et al., 1961), where they are measured at our station. In winter, the wind speed profile shifts to the right, peaking later in the day at around 14:00 UTC. This shift to later in the day occurs because, in winter, it takes more time to reach the required thermal contrast between the land and the sea to form sea breezes. Note the existence of a second peak in the wind speed plot during the night, around 01:00 UTC, representing the land breezes. These land breezes are stronger in winter than in the other seasons.

The time profiles of air temperature (Fig. 2b) are relatively flat, showing a weak diurnal cycle. Winter reveals the most pronounced diurnal cycle. The temperature contrast between day and night is minimal in summer and maximal in winter. The weak diurnal cycle observed in the station-measured temperature is because of the influence of SST, since the station is located very close to the sea. The diurnal cycle of land temperature becomes much stronger as we go further inland in the coastal region (Fig. S1 in the Supplement), creating a strong temperature gradient between the ocean and the land surface, which ultimately drives the breeze circulation.

Given the strong diurnal cycles of surface winds and temperature, it is evident that the day and night circulation in the study area is remarkably different. Therefore, it becomes important to look at the aerosol vertical profiles separately in the day and night.

Figure 14 shows the diurnal cycles of aerosol extinction in KAUST–MPL data across the entire atmospheric column. Dust is generally confined within the lowest ∼2 km in winter and reaches ∼6 km in summer, following the seasonal and diurnal variations of the PBL. Note that there are some gaps in the KAUST–MPL data because of the quality controls applied. In summer, there is significant dust activity in the morning (∼06:00 local time), and in spring, dust activity peaks throughout the afternoon. In winter, the KAUST–MPL shows more vigorous dust activity in the nighttime (21:00 to 00:00 local time) near the surface. This increased dust activity at night is due to the effect of land breezes, which are strongest in winter (Fig. 2). We explore the effect of breezes on dust emissions and transport in Sect. 3.5. KAUST–MPL shows high extinctions at a height of 6–7 km, particularly in the evening of summer and fall, which represent long-range transported dust during large-scale dust events. Such high-intensity dust events are more frequent in summer and fall, as observed in the KAUST–MPL data (Figs. 10 and 11).

Figure 15 shows the circulation features of land and sea breezes in the vicinity of the KAUST–MPL site. The base map in the figure shows the high-resolution dust source function used in this study, where red hotspots represent the most dominant dust sources. Significant dust sources are observed on both sides of the Sarawat mountain range, i.e., the coastal sides and the eastern slopes. Sea breezes are strongest in spring and summer. In contrast, land breezes are strongest in winter and fall. In the daytime, sea breezes penetrate further inland, and the KAUST–MPL site receives northwesterly winds. At night, the KAUST–MPL site experiences northeasterly land breezes, which are strongest in winter.

Figure 16 shows the total aerosol concentration (µgm-3) within the innermost model domain (d03) in a longitudinal cross section perpendicular to the coastline over KAUST. The section also shows the land profile (black shades) where the Sarawat Mountains that run along the eastern coast of the Red Sea and the relatively flat inland deserts that lie on the eastern side of the mountains are visible. The mountains reach a maximum elevation of ∼1.5 km above sea level. The effect of land and sea breezes on dust is apparent in Fig. 16, as discussed in further detail below.

During winter nights, a thin layer of dust collects over the marine boundary layer and the land near the KAUST site within ∼1 km height. This layer is an accumulation of dust that has been mobilized by land breezes from the coastal plains and the western flanks of the mountains. The coastal plains of the Red Sea are rich in fine fluvial sediments deposited by wadis, which are known sources of dust (Anisimov et al., 2017; Parajuli et al., 2019). The western flanks of the mountains also contain fluvial and intermountain deposits along the slope that are suitable for resuspension (Parajuli et al., 2014). This mobilized dust is transported towards the Red Sea, which seems to occur at low altitudes ∼500 m (Fig. 16). Some dust collects over the Red Sea during the daytime in the winter also, which appears well mixed within the relatively shallow PBL. During the day, the northwesterly sea breezes move landward, preventing the dust emitted from the coastal region from moving over the sea. Therefore, this dust observed during the daytime must be the residual dust that accumulated overnight. The dust mobilization from the coastal area by the sea breezes (daytime) is weaker during the winter.

In the spring, there is very high dust loading over the coastal region and the western flanks of the mountains, which is much higher than in winter. This higher dust loading is consistent with stronger sea breezes in spring than in winter (Fig. 15). The highest dust loading is observed over the slopes of the mountains at a height of 1–1.5 km. Recall that the lidar data show a high dust loading at ∼1–1.5 km height at the KAUST site. Two factors appear to contribute to this high dust loading. First, daytime sea breezes mobilize dust locally from the coastal plains and the western flanks of the mountains. These sea breezes then push the dust inland and upwards along the slope of the mountains, up to 3 km height. At the same time, the northeasterly Harmattan winds also bring dust from the nearby inland deserts towards the mountains. This dust is further uplifted when the dust-laden Harmattan winds encounter the sea breezes coming from the opposite direction. Thus, the interaction of sea breezes with the northeasterly Harmattan winds across the mountains mainly determines the vertical distribution of aerosols over the region. At night, the sea breezes as well as the PBL weaken, and the vertical extent of dust in the atmosphere reduces. The land breezes also appear to transport the dust towards the Red Sea from the western flanks of the mountains at night.

In summer, the patterns of dust mobilization and transport are similar to those in spring but are not quite as pronounced. In fall, the mobilization of dust from the coast and its ocean-ward transport is very weak, and the patterns are similar to those in winter.

Figure 17 shows the daytime and nighttime winds at three altitudes for two specific months in summer (August) and winter (February). Note that the winds are shown at different levels for August and February to highlight the features of land and sea breezes better. The depth of sea breezes and land breezes are different, as expected, with the sea breezes being much deeper than the land breezes, primarily because the PBL is higher during the day than at night. The local topography also plays a role. Sea breezes are still strong up to a height of ∼1150 m; however, the land breezes only reach a height of ∼200 m. By about 450 m, the land breezes subside completely. The land breeze circulation is confined by the height of the mountains, whereas the sea breeze circulation extends to a much higher altitude. The returning flow of the sea breezes takes place at a height of ∼2250 m in the form of northeasterly trade winds, which are responsible for bringing the dust to our site from the inland deserts. The return flow of the land breezes occurs at a height of ∼1500 m, with a change in direction of nearly 180∘ of the lower part of the subtropical westerly jets (de Vries et al., 2013) (see supporting information in Fig. S6). The variation in the pattern of these winds along the vertical dimension is generally consistent with the profile of modeled dust that we presented earlier (Fig. 16).

In summary, the timings and patterns of dust emission and transport in the study region are evidently affected by land and sea breezes. Note that, across the larger parts of the Arabian Peninsula, the seasonality of dust mobilization is quite different to our study region, where dust emission and transport are maximal during summer (Parajuli et al., 2019).

The model simulated the surface wind speed at the KAUST site reasonably well as compared to station data (Fig. 3). Accurately representing the surface winds is vital because the dust emission is parameterized as a function of friction wind velocity in WRF-Chem (Marticorena and Bergametti, 1995; LeGrand et al., 2019). Note that dust emissions are generally caused by wind gusts that occur over very short timescales (seconds) (Engelstaedt and Washington, 2007), which are much stronger than the average seasonal wind speed displayed in Fig. 2. We can expect these wind gusts to be represented in our simulations because we have used a very small model time step (8 s) in our d03 domain. Given our primary focus is on vertical aerosol profiles, further analysis of wind gusts is beyond this study's scope.

The model reproduces the AOD time series well in all seasons as compared with several datasets, including AERONET, MODIS, MISR, and MERRA-2 (Fig. 4), with an MBE of 13.4 % against AERONET data. There is some mismatch in the AOD profiles among different datasets during some large-scale dust events, partly because of the difference in sampling and measurement frequencies.

The model successfully captured the evolution of a dust event that occurred in 2015 over the study site in terms of its onset and demise, as well as the height of the dust layer (Fig. 9). Our results were consistent with several previous studies, such as in Yuan et al. (2019) and Anisimov et al. (2018). The model generally reproduced the elevated dust layers at ∼6 km during the dust event (Fig. 9), which were prominently seen in KAUST–MPL observations (Figs. 10 and 11). However, the model underestimated the AOD at KAUST by about 35 % during the event compared to AERONET AOD. Simulating these complex, large-scale dust events is extremely challenging, and thus we do not expect the model to capture them as precisely, since they occur only a few times (∼2–3) in a year. We note that the performance of WRF-Chem in simulating these large-scale dust events is case-specific (e.g., Teixeira et al., 2016; Fernández et al., 2019) and should not be generalized. The model performance was indeed sensitive to the type of dust event (e.g., Kim et al., 2017), the details of the dust-emission processes (Klose and Shao, 2012; Klose and Shao, 2013), the dust source function used (Kalenderski and Stenchikov 2016; Parajuli et al., 2019), and the prescribed size distribution of the emitted dust (Kok et al., 2017; Marenco et al., 2018).

The seasonal aerosol vertical profiles were consistent among all datasets that we compared viz. KAUST–MPL, MERRA-2, and CALIOP (Figs. 10, 11 and 13). These results are consistent with those reported by Li et al. (2018) over the same region. The WRF-Chem model successfully reproduced the vertical profiles of dust aerosol extinction and concentration in terms of seasonality when compared with the abovementioned datasets. Nearer the surface, the model showed some disagreement with the observational datasets, as also noted in some previous studies (e.g., Hu et al., 2016; Wu et al., 2017; Flaounas et al., 2017). Note that such disagreement between data collected near the surface exists among the observational datasets as well; this disagreement could arise due to differences in the retrieval algorithms used and the resolution of the datasets, as discussed in detail below.

The difference in vertical profiles retrieved from KAUST–MPL and CALIOP data could be related to the differences in the algorithm and resolution between the two datasets. Firstly, while retrieving aerosol extinction profiles, the CALIOP algorithm uses different prescribed extinction-to-backscatter (lidar ratio) for a set of aerosol types from a lookup table (Omar et al., 2009; Winker et al., 2009; Kim et al., 2018). In addition, the CALIOP algorithm has difficulty in identifying the base of aerosol layers accurately. In particular, the level-3 algorithm ignores the “clear air” between the surface and the lowest aerosol layer when averaging to avoid underestimation of extinction in the lower part of the aerosol profile (Winker et al., 2013). In contrast, the MPL algorithm assumes an averaged lidar ratio for the whole column based on the aerosol PSD, refractive index, and sphericity, in such a way that it satisfies both AERONET and MPL co-incident data. Because of the assumption of a constant lidar ratio, MPL retrievals near the surface could be erroneous, especially when multiple aerosol layers are present (Welton et al., 2002a). Secondly, KAUST–MPL is a point measurement that captures the temporal evolution of the dust storms better than CALIOP because it has a higher temporal resolution. For instance, CALIOP can undersample or overlook some dust events that last only for a few hours. On the other hand, CALIOP could sample more spatial details of a dust storm because of its extended coverage along its track compared to KAUST–MPL data. Nonetheless, these two datasets complement one another, and their combined use can be beneficial in understanding the large-scale dust storms.

Analysis of the KAUST–MPL data revealed several interesting features of the vertical profile of aerosols over the study site, which were not documented in earlier studies. For example, we observed a significant difference between the daytime and nighttime vertical profiles of aerosols. Some of these detailed features were not apparent in the model simulations. The model underestimated the nighttime aerosol extinctions at ∼6–7 km height in summer and fall compared to the KAUST–MPL data (Figs. 10 and 11). Although the model data did not identify these dust layers at 6–7 km in the seasonally averaged profiles, the model nonetheless correctly identified these same dust layers during the dust event analyzed in the case study (Fig. 9). This result supports our speculation that the elevated dust layers at ∼6–7 km represent transported dust from inland deserts during large-scale dust events.

It is difficult to identify the exact reason for the above discrepancy between the model and KAUST–MPL data, but there are several possible explanations. First, the model could be deficient in representing the deep convective mixing of dust in the central-peninsula deserts. Second, although the effect of orography on dust seems to be correctly resolved (Fig. 16), the long-range transport of dust from the deserts towards the KAUST site may not be fully detected. Third, part of this discrepancy could also be because of the insufficient model spatial resolution compared to KAUST–MPL data. KAUST–MPL data are a point measurement, while the model data represent the profiles at a 1.3×1.3 km grid cell, which, although high resolution, can still produce a substantial difference, especially in a land–ocean boundary. Finally, the discrepancy could also be due to the limitation of the GRASP algorithm in handling clouds, because of which the aerosol layers observed at 5–7 km height in the nighttime could be contaminated with clouds, as explained further below.

To better understand the origin of two elevated dust layers observed (∼1–2 and 6–7 km) and investigate the possibility of thin-cloud contamination in our MPL retrievals, we analyzed the volume-depolarization profiles provided by the KAUST–MPL, synchronous with the attenuated backscatter profiles used in the retrievals. The average volume depolarization value in the lower atmosphere (1–2 km) was estimated to be 13 %–14 % on average and 7 %–8 % for the upper part (6–7 km) for the selected period. Such values indicate that high extinction values in this altitude range cannot come exclusively from clouds because pure water clouds generally yield a 1 %–2 % depolarization value and ∼30 % or even higher in the case of cirrus clouds (e.g., Del Guasta and Valar, 2003). The lower depolarization value in the upper part could be explained by the fact that the aerosol particle sizes are much finer than those in the lower part. At the same time, a lower depolarization value also suggests the possibility of partial influence by thin clouds. The presence of thin clouds can probably cause some overestimation of aerosol concentrations and extinction at these altitudes. However, such an overestimation is expected to increase the fitting errors, which are easily detectable, as mentioned earlier. To ascertain this with full confidence, we plan a further analysis utilizing simultaneous retrieval of Sun-photometric observations together with backscatter and volume-depolarization profiles provided by KAUST–MPL in the future.

Although both model results and the KAUST–MPL retrievals have their own limitations, both KAUST–MPL and the model data identified two prominent layers of dust over the study site, one at a lower altitude (∼1–2 km) and another at a higher altitude (∼6–7 km). These two dust layers correspond to two different dust sources. The lower dust layer corresponded to dust originating from nearby deserts, and the upper dust layer corresponded to dust coming from more remote sources and further inland. The two layers of dust are typical in this region during dust events. As explained before, a large-scale disturbance usually brings dust from remote sources at higher altitudes (∼6–7 km). When the disturbance comes closer to the site, high surface winds associated with the disturbance also pick up more dust from nearby deserts giving rise to a high dust loading at ∼1–2 km height. Such stratified aerosol layers have been previously observed near land–ocean boundaries, where strong temperature inversion occurs, restricting further mixing of aerosols in the PBL and above (Welton et al., 2002b).

In the lower part (∼1–2 km), the atmospheric dust loading is mostly dominated by coarse-mode particles. In contrast, dust in the upper level (∼6–7 km) typically constitutes long-range transported finer particles. Finer particles can easily reach the upper atmosphere, whereas coarser particles of higher mass fall back to the surface more quickly due to gravitational settling. Thus, coarser particles are usually confined to the lower atmosphere, have shorter atmospheric lifetimes (∼1–3 d), and affect hourly/daily scale climate processes such as the diurnal cycle. On the contrary, smaller particles reach higher altitudes and have longer atmospheric lifetimes. The extinction cross section of an individual large particle is bigger than that of a small particle, but finer particles have stronger radiative effects per unit mass than coarser particles (Khan et al., 2015).

We observed some interannual variability while comparing the vertical profiles for 2015 and 2016, but was not too significant (Fig. S9). Therefore, the observed vertical distribution of dust aerosols can be considered “typical” for our region and possibly for other land–ocean boundaries (e.g., Rasch et al., 2001). This is understandable because the synoptic winds causing large-scale dust events, and the diurnal-scale breezes that affect the dust distribution, both have strong seasonality over the study region (Kalenderski and Stenchikov, 2016; Parajuli et al., 2019). However, as demonstrated by our results, vertical profiles of aerosols can be affected by regional processes such as breezes, which indicate that the profiles can differ across different regions. Therefore, it is vital to examine the aerosol vertical profiles of a region to understand the regional climate.

KAUST–MPL-retrieved aerosol vertical profiles also provided an opportunity to understand how aerosols interact with land and sea breezes over the eastern coast of the Red Sea. The salient features of the land and sea breezes over the study region revealed by our study are summarized in Fig. 18. These fine-level interactions are often poorly resolved in coarse-scale simulations. Our high-resolution simulations (∼1.33×1.33 km) nonetheless correctly resolved these features and showed how breezes affect dust aerosol distribution over the region. Our study is important because the breezes and dust can directly affect the daily life of populations that reside in the coastal area. Furthermore, dust over the region affects the surface temperature of the Red Sea through changes in radiation (Sokolik and Toon, 1996; Osipov et al., 2015; Osipov and Stenchikov, 2018), which could have an enormous impact on the Red Sea climate and marine habitats. Additionally, changes in dust deposition also affect the availability of nutrients delivered to marine ecosystems (Jish Prakash et al., 2015).

MPL data are invaluable for studying the vertical details of aerosols in the atmosphere because they measure backscatter from aerosols and clouds with a high vertical and temporal resolution (Welton et al., 2002b; Winker et al., 2009). Most satellite data only provide aerosol properties over the entire atmospheric column (e.g., Hsu et al., 2004), which are complemented by the MPL data that provide height, depth, and the particle characteristics of the aerosol layers in the atmosphere. Since satellite data usually have a low temporal resolution and, because many large-scale dust events are short-lived, MPL data can reveal additional characteristics of dust storms.

In regional and global climate models, it is a usual practice to constrain the total AOD using some observations (see, for example, Zhao et al., 2010; Parajuli et al., 2019). While such constraints are desirable because they help to represent columnar atmospheric properties more precisely, they are not sufficient for certain applications such as air quality modeling, for example (Ukhov et al., 2020b). Unless the model correctly represents the aerosol vertical profiles, the model-estimated surface aerosol concentrations may not be reliable. In this context, KAUST–MPL data can be instrumental in constraining the vertical distribution of aerosols in the models. Such constraints would ideally benefit the operational forecasting of dust storms and air quality (Zhang et al., 2015).

Although derived from actual observations, KAUST–MPL retrievals are also subject to uncertainties, and their accuracy is dependent on assumptions made by the retrieval algorithms. A study that compared the GRASP retrieval scheme employed here against in situ measurements showed that the differences were less than 30 % for the different retrieval schemes (Benavent-Oltra et al., 2019).