Desert dust aerosols have a substantial influence on the atmospheric environment of the Middle East e.g. and on the maritime environment of the Red Sea e.g.. The latter is pinned between two of the largest sandy desert regions in the world, with the Sahara of northern Africa to the west and the Arabian and Syrian deserts to the east. Of broader scientific, environmental, and cultural interest in this region is the biodiversity of the Red Sea and its substantial coral reef systems, which are at risk of increasing global ocean temperatures and which have experienced substantial degradation and “bleaching” in recent times e.g.. Over land, the Middle East and the Arabian Peninsula have seen a recent increase in dust activity over the past 10 years , a situation aggravated by drought conditions in the semi-arid regions of Syria and Iraq in the north (the Levant). It is an open question as to whether this increase in activity has propagated through to increased atmospheric dust loadings over the Red Sea, and similarly over the Persian Gulf to the east of the Arabian Peninsula.

Despite the region's vulnerability to the impacts of desert dust aerosol, it is relatively little monitored compared to the Mediterranean to the north and to the neighbouring desert regions. The Sahara, for example, has been investigated over the past 10 years by numerous field campaigns such as AMMA, SAMUM, GERBILS, and Fennec (; ; ). Aerosol Robotic Network (AERONET) sunphotometers , which are part of a global network aiming to provide long-term observations of atmospheric aerosol loading, are currently only located on the Red Sea coast at Eilat in Israel (at the far north of the Red Sea) and the King Abdullah University of Science and Technology (KAUST) in Saudi Arabia at 22.3∘ N, and the latter was only established in February 2012.

Nevertheless, (subsequently denoted B15) recently investigated the performance of satellite retrievals over the Red Sea, making use of ship cruise data taken under the framework of the Maritime AERONET Network . They found close agreement between dust aerosol retrievals from the Spinning Enhanced Visible and Infrared Imager (SEVIRI) and the MODerate resolution Imaging Spectroradiometer (MODIS) instruments. SEVIRI-MODIS aerosol optical depth (AOD) offsets were found to be +0.02, with correlations of 0.93 and 0.95 for the two sets of cruises considered, with similar statistics obtained when considering the agreement between the two sets of retrievals and the ship-based measurements. The same study identified a strong seasonality in dust loading over the sea and a particularly marked latitudinal gradient in summertime dust AOD. This differential loading was shown to propagate to significantly enhanced surface radiative cooling in the southern part of the basin relative to the north.

However, the B15 study was limited to a 5-year period from 2008 to 2012 and used MODIS Collection 5 data, which have since been superseded by Collection 6 . More critically, the ship-based cruise data used to evaluate the retrievals did not sample conditions typical of the summertime southern Red Sea. identified systematic differences between dust retrievals over the Sahara, influenced by such factors as dust loading, atmospheric moisture, and surface emissivity. It is to be expected that, over a longer time period and given sufficiently high dust loadings, the retrievals over the Red Sea may also exhibit systematic differences.

Hence, in this paper, we extend the B15 analysis and investigate in greater detail the latitudinal structure of dust aerosol presence over the Red Sea, as inferred from SEVIRI and MODIS Collection 6 retrievals. The 11-year pattern in dust activity is presented and discussed, with reference to both the temporal and spatial structure. We make use of all the ship-based cruises now available over the Red Sea (from 2010 to 2015) and the KAUST AERONET measurements to evaluate the SEVIRI and MODIS AOD retrievals. To deepen our understanding of the differences that we find between the SEVIRI and MODIS retrievals, we also make use of AOD measurements from the Multi-angle Imaging SpectroRadiometer (MISR) satellite instrument, and explore the performance of the retrievals over the Persian Gulf in comparison with that over the Red Sea.

Examining retrievals over sea/ocean only, the “wider” region investigated here is within the bounds of 12–30∘ N, 32–56∘ E, encapsulating both the Red Sea and nearby waters including the Persian Gulf and parts of the Arabian Sea. This wider region is used to set activity over the Red Sea into context. Meanwhile, the Red Sea itself (referred to as “the Sea” hereafter) is located within the region bounded by 12–30∘ N, 32–44∘ E. Egypt, the Republic of the Sudan, Eritrea, and Djibouti border it to the west, Israel and Jordan to the north, and Saudi Arabia and Yemen to the east. The southernmost point at 12.58∘ N is the Bab-el-Mandeb, a narrow strait between Yemen and Djibouti. At the northern end, the Red Sea again narrows considerably where it splits into two, the gulfs of Suez (west) and Aqaba (east), ending at 29.97∘ N. The total length of the Red Sea from the Bab-el-Mandeb to the northern end of the Gulf of Suez is approximately 2200 km, but the maximum width of the Sea is less than 400 km. This narrow width of the Sea is such that it can effectively be subdivided into latitude bands to describe the various subdomains within the basin. A map of the region and its topography is presented in Fig. . The Persian Gulf is a similarly nearly enclosed sea which is susceptible to dust activity and is located to the east of the Arabian Peninsula; in the context of this study, it is bounded by the longitudes 48 and 56∘ E, and the latitudes 24 and 30∘ N.

An advantage of the duration that MSG-SEVIRI and the MODIS instruments have now been in orbit for is that we can set the inter-annual variability of dust activity into context, and extend the 5-year record presented by B15. It has been noted before that the southern part of the Sea is often dustier than the north, for example by B15, and we extend this previous work by investigating the latitudinal variations in dust loading over the Sea over a longer time period. Figure  shows a time series of monthly mean SEVIRI AODs at 630 nm over the Red Sea from 2005 to 2015. The Sea is here subdivided into north and south, divided by the line of 20∘ N which splits the Sea into very closely equal regions by area (13 725 pixels in the north, 13 915 in the south).

The annual cycle is clear and consistent: dust activity peaks in summer, particularly in July, and is concentrated in the southern half of the Sea. The north has an annual cycle, generally peaking in summer, but the amplitude is much less pronounced than it is in the south. Looking for exceptions to the general pattern, 2015 is a somewhat unusual year in that there is no one dominant peak; instead, dust activity is spread between the months of April, June, and August. In general, the annual cycle is as described in Fig. 5 of B15, but we see further interesting features in the breakdown into halves of the Sea and in the extension of the record. Looking at the southern end of the Sea, 2005–2007 is a relatively quiet period, transitioning during 2008 to the dusty period of 2009–2013. This may be associated with the consequences of the Levantine drought period which resulted in enhanced dust activity over the Arabian Peninsula from 2008/2009 onwards . Latterly, however, the enhanced summer dust activity seems to have tailed off in 2014 and 2015.

Only in July 2010 and August 2015 does the mean AOD in the north exceed 0.5. In the south, dust flows can be trapped above the Sea due to the mountainous topography to either side in both Africa and the Arabian Peninsula (see Fig. ). By contrast, in the north, the relatively flat topography of the land to either side is less of a hindrance to dust transport, and hence dust has less of a tendency to persist for extended periods of time in the atmosphere above the northern part of the Sea.

Spatially, on the decadal timescale, as indicated in Fig. , a sharp gradient in dust loading across the Sea is most strongly apparent in summer (represented here by July). This is also the only season when there is any significant dust activity over the Arabian Sea. For the most part, this timescale disguises specific geographical features. An exception to this is that dust outflows into the southern Red Sea from the Tokar Gap in the Republic of the Sudan e.g. are readily apparent in summer (Fig. c), just south of the 20∘ line of latitude. As well as being a dust source, the mountain-gap wind jet in this location during summer affects the eddy mixing and circulation of the Sea itself . Further south, there are more dust sources on the Eritrean coastal plain which contribute to the high southern dust loading, but in general much of the dust actually originates from the central Arabian Peninsula and the Republic of the Sudan.

Cloud cover only tends to have a significant effect on the retrieval quantity in summer for one or two months around July during the monsoon period e.g., when there is persistent cloud presence in the south, coincident with the typical maximum of dust activity; the fraction of successful retrievals with respect to the total number of attempted retrievals over the southern half of the Sea can be as low as 20 %. Cloud cover in the north is typically quite sparse year round.

In order to verify these patterns in the SEVIRI dust observational record over the Red Sea, it is vital to validate and compare these retrievals against other data sources such as recent ship cruise data. From October 2010 to September 2015, multiple research ship cruises organised by KAUST have taken place on the Red Sea, principally based out of KAUST itself, and Georgiy Stenchikov is the PI of the ship-based aerosol observation campaign. Aerosol measurements were made using a handheld Microtops sunphotometer, carried out following the framework of the Maritime AERONET of ship-based aerosol measurement campaigns . B15 carried out validation of SEVIRI and MODIS retrievals using these data, but only for 2011 and 2013, and only for MODIS Collection 5 data. We repeat this process here using the updated Collection 6 data and including 2010, 2012, 2014, and 2015. As of the end of 2015, there were 23 ship cruise legs available in the Red Sea at Level 2 quality. Otherwise, the same co-location criteria were used: all three instruments must make a successful retrieval, averaged for satellite pixels within a 50 km radius of the ship and for retrievals within an hour to either side of the sunphotometer measurement. As stated earlier, we only use MODIS AOD retrievals with QA values from 1 to 3. For consistency with the SEVIRI retrievals, the AERONET and MODIS AOD values are scaled to a wavelength of 630 nm using the retrieved Ångström exponent. B15 found 111 co-locations; over the expanded dataset for the Red Sea, we now find 255 co-locations. Note an imbalance in the number of measurements between the northern (> 20∘ N) and southern parts of the Sea: there are 220 co-locations in the north, compared to just 35 in the south. Due to political considerations, it was more difficult for the ship cruises to spend much time at the southern end of the Sea. Validation statistics are listed in Table .

During this period, few major dust events have been observed from the ship-based measurements over the Sea. The co-located maximum measured AOD at 630 nm by the sunphotometers at sea level was 0.86 in July 2011, at a latitude of 26.3∘ N. The mean AOD measured was 0.28 with an associated standard deviation of 0.15. The sea-based measurements correlate with both the SEVIRI and MODIS AODs to a value of 0.92. Meanwhile, the respective satellite retrieval biases with respect to the ship-based measurements are +0.010 for SEVIRI and +0.029 for MODIS. The RMSDs are 0.059 for SEVIRI and 0.066 for MODIS. Thus, the agreement between all three data sources is good, although there is a tendency for the MODIS Dark Target AODs to be positively biased against both the sea-based and the SEVIRI data. The correlations and RMSDs are broadly in line with previous statistics calculated by B15; however, the previous work suggested a slightly higher positive bias by SEVIRI against MODIS, indicating that the average MODIS AODs may have nudged slightly higher (by ∼ 0.01–0.02) from Collection 5 to Collection 6.

A similar picture is present when we look at AERONET sunphotometer data taken at the KAUST campus (22.31∘ N, 39.10∘ E), here analysed for the years 2012–2015. This site has also been established and organised by KAUST, with Georgiy Stenchikov as the PI (http://aeronet.gsfc.nasa.gov/new_web/photo_db/KAUST_Campus.html). Scatterplots of satellite retrieval AODs against AERONET AODs are shown in Fig. a and c. The Abu Al Bukhoosh site (panels b and d, at 25.50∘ N, 53.15∘ E) will be discussed in further detail in Sect. . A caveat for the KAUST site is that we are comparing AERONET measurements taken over land, albeit very close to the shoreline, with satellite retrievals taken over sea. The same spatial matching criteria are applied as for the ship-based measurements, although due to the higher frequency of observations on land we restrict the temporal matching to within half an hour on either side. Level 2 AERONET data are used for KAUST, which are cloud-screened following the technique described by . Given the urban/industrial environment at Jeddah ∼ 80 km down the coast from KAUST, we also apply the criterion that the AERONET Ångström coefficient must be less than 0.6, in order to distinguish dust as being the dominant aerosol e.g.. Over the 4 years considered, there are 595 points when all of the three instruments made a simultaneous successful retrieval over KAUST, in which the AERONET measurements had a mean AOD of 0.47 ± 0.34. The SEVIRI retrievals were correlated with AERONET to a value of 0.96, and MODIS was correlated with AERONET to a value of 0.92. Meanwhile, the respective biases were 0.000 and +0.080, and the RMSDs were 0.095 and 0.210. The MODIS RMSD is particularly high due to a handful of points at AERONET AODs > 2 where MODIS retrieves AODs > 4 (Fig. a), which contributes to MODIS' high positive bias. Hence, while the overall picture is consistent with the ship-based measurements, there seem to be higher RMSDs when sea pixels near the coast are considered, perhaps due to the higher mean AODs but perhaps also due to the higher variability in the sea-surface reflectance in coastal waters as opposed to the open ocean. Another possible factor is the extra distance between the site and the satellite-retrieved pixels given that the retrievals avoid pixels over land and coastal waters.

From Figs.  and , it is apparent that the climatological monthly summertime SEVIRI AOD over the Red Sea often exceeds 1, a range not encapsulated by the ship cruise measurements. It is important to check the agreement of the SEVIRI AODs with other retrievals such as MODIS in this regime, given that the higher the AOD, the larger the impact of dust will be on the radiative energy budget at the surface, within the atmosphere and at the top of the atmosphere. Figure  adapts Fig.  to compare MODIS AOD retrievals with the coincident SEVIRI retrievals (averaged at 07:00, 08:00, and 09:00 UTC for Terra and at 10:00, 11:00, and 12:00 UTC for Aqua), along with statistics of the inter-retrieval offset and root mean square difference. The approximate daily overpass times are 08:00 and 11:00 UTC for Terra and Aqua over the longitude of the Red Sea; averages of the SEVIRI retrievals are taken on either side of the main overpass times in order to provide successful retrievals when sun glint may prevent SEVIRI retrievals, as is the case for coincident Terra overpasses in the morning in midsummer. These monthly statistics are calculated by investigating the coincident and binned 0.125∘ MODIS and SEVIRI grid cells, when and where both instruments make a successful aerosol retrieval. Note that the original MODIS AOD retrievals at 550 nm have been scaled using the retrieved Ångström coefficient to 630 nm in order to match the wavelength retrieved by SEVIRI.

For almost all months, the MODIS retrievals have higher values than the SEVIRI retrievals, with the largest values of offsets and RMSDs when the dust loading is largest. Consequently, the offsets and RMSDs tend to be larger in the south of the Sea than in the north. The correlations are also slightly worse in the south: the average monthly correlation is 0.93 in the north and 0.91 in the south. MODIS-SEVIRI offsets at the basin scale can be much larger than at the ship-point scale, exceeding +0.3 in the southern Red Sea for the July periods of 2009, 2011, and 2013. A caveat to this general pattern is in the autumn to early-spring months when the mean AOD is low, and there are months when there is a very slight negative MODIS offset. Given this behaviour, the latitudinal gradient in the AODs is also readily apparent in the MODIS-SEVIRI comparison statistics (Fig. c, e). The AOD offsets and RMSDs are quite negligible in autumn and winter (below ±0.05 of offset and below ∼ 0.1 of RMSD), but show much more variability in July in the south, when there is the most variability in the intensity of large dust outbreaks. Outside of the summer months, retrievals over the southern part of the Sea in April 2015 also show large deviations, a reflection of the relatively high dust loadings seen in this month and location.

The discrepancy between MODIS and SEVIRI at high AODs is systematic, as can be seen in Fig. , which considers all of the co-located SEVIRI and MODIS retrievals over the 11-year period. Comparing MODIS against SEVIRI, the correlation coefficient exceeds 0.9 for all months, with the RMSDs and offsets peaking when the AOD is highest. Strikingly, the 2-D histogram curves further away from the 1:1 line as the AOD increases, as SEVIRI appears to become less sensitive to the increases in AOD that MODIS seems to observe. Hence, in April and July, when the tail of high AODs is extended, the two retrieval datasets diverge the furthest. Very similar patterns are seen in comparisons between Terra-MODIS and SEVIRI, and Aqua-MODIS and SEVIRI (not shown), indicating broad consistency between the separate MODIS instruments.

An implication of this is that there are two regimes for the MODIS/SEVIRI comparisons, which can be seen more precisely if we compare MODIS AODs against the MODIS-SEVIRI offset. At low AODs < 1, the MODIS-SEVIRI offset is small, and neither instrument retrieval is consistently reporting higher AODs than the other. Consequently, the correlation between the MODIS AOD and the MODIS-SEVIRI offset in January and October is weak, at 0.47 and 0.41 for the respective months. In spring and summer, however, there is a marked increase in the correlation between MODIS and its offset against SEVIRI, increasing to 0.83 and 0.85 in April and July.

Examining the MODIS-retrieved Ångström coefficients over this period provides more detail as to what is occurring here. At low AODs < 1, the mean Ångström coefficient is 0.67, while at AODs > 1, the mean is 0.14. The low value of the latter is a clear signature of desert dust , given that sea salt alone would not result in such high AODs. Hence, it is clear that the substantial majority of the points at the tail end in Fig. b and c are composed of desert dust, and so it is dust in some form which is responsible for the high-AOD discrepancy between the SEVIRI and MODIS retrievals. Contamination by other aerosol types is much more likely at lower AODs, when urban/industrial pollution from the major centres along the Red Sea coast (e.g. Jeddah, Port Sudan) may contribute more to the overall signal.

This positive offset behaviour exhibited by MODIS with respect to SEVIRI leads naturally to the question of which retrieval is more accurate at the highest AODs. It is difficult to assess this using data from the ground or from the sea surface given the paucity of measurements made at the surface of such intense events. The ship-board sunphotometers only measured a maximum AOD of 0.86, while the AERONET site at the KAUST campus has only made co-located measurements of AODs > 1 for 42 points out of 595 (7 %). Within this subset of high-AOD data, the MODIS bias with respect to AERONET is +0.313, while the SEVIRI bias is -0.075; this is in comparison to the biases at low dust loadings of +0.063 for MODIS and +0.006 for SEVIRI. Hence, as the dust loading increases, so does the MODIS bias with respect to the AERONET validation dataset while SEVIRI becomes slightly negatively biased. However, this is a relatively small subset of cases from which it is dangerous to draw any substantiative conclusions. With this in mind, we introduce the use of MISR aerosol retrievals to investigate whether they show more coherent behaviour with one of the MODIS or SEVIRI sets of AOD retrievals. It is important to note that MISR has quoted uncertainties of equivalent magnitude to the MODIS and SEVIRI retrievals (see Sect. ), and so they are included here simply for another point of comparison.

As previously reported in comparisons between MISR and Terra MODIS Collection 5 retrievals for all aerosol types over the global ocean for January 2006 , MISR retrieves higher AODs than MODIS when the AOD is low, whereas when the AOD is greater than 0.2–0.3 the MODIS AODs become “systematically” higher than the MISR AODs. A similar picture can be found over the Red Sea when we look at scatterplots of MISR/MODIS comparisons for the months of January, April, July, and October for the years 2008–2011 (Fig. b). In this case, the datasets have been gridded at 0.25∘ resolution to take into account the 17.6 km spatial resolution of the MISR aerosol product. These have been co-located with successful SEVIRI retrievals, with the MODIS and SEVIRI AOD datasets reported at, or scaled to, 550 nm, while the MISR AODs used are reported at 558 nm. MODIS and SEVIRI have been retrieved from Aqua or at the Aqua time, due to persistent sun glint affecting both instrument retrievals along the MISR track at the Terra morning time during the spring and summer months. This persistent sun glint issue has been noted before by , in their Fig. 4c, which indicates consistent failure of co-located AOD retrievals between MISR and Terra-MODIS in the northern tropics in July.

The temporal discrepancy accounts for much of the scatter in Fig. , especially in July. At lower AODs in the winter months, there is more temporal stability in the daytime AOD values and hence less scatter. In all 4 months, the overall offset of AOD retrievals from Aqua-MODIS is negative when compared to MISR. However, consistent with this masks contrasting behaviour in different AOD regimes which is most obviously manifested in July, the month which has the smallest overall offset. Over the whole dataset, positive offsets at higher AODs (+0.188 when MISR AOD > 1) are obscured by negative offsets at lower values (-0.045 when MISR AOD < 1).

Taking MISR AOD retrievals as our reference, and benefitting from its multi-angle viewing capabilities e.g., Fig. a indicates that SEVIRI also has a negative offset as compared with the MISR retrievals but that this offset is consistently in one direction regardless of the MISR AOD. Using a MISR AOD equal to 1 as an arbitrary cut-off between high and low dust loading, SEVIRI has a negative offset of 0.062 (high AOD) and of 0.071 (low AOD) relative to MISR. Thus, the discrepancy between regimes is much reduced compared to MODIS. While it is impossible to state which, if any, of the retrievals are correct, it is clear that there are substantial, and in some cases obviously systematic, differences between all of the retrievals when elevated dust optical depths are seen over the Red Sea.

There are a number of possible explanations for the discrepancies seen between MODIS and SEVIRI at high AODs. The aerosol model used in the SEVIRI retrievals assumes a purely scattering aerosol. While measurement campaigns indicate that dust typically has a high single-scattering albedo (SSA) in the visible, some absorption is always observed. For example, aircraft measurements during the GERBILS campaign in June 2007 indicate SSA values between 0.92 and 0.99 ; more recent values derived from Fennec campaign measurements indicate rather lower values, ranging from 0.70 to 0.97 . Hence, one might anticipate that SEVIRI would retrieve a smaller AOD than that which would be obtained using a model with a lower SSA, all other aspects remaining the same. Similarly, another contributor to the discrepancy may be the different spectral ranges used by the retrievals: while SEVIRI only uses the 630 nm channel for this AOD dataset, the MODIS ocean AODs are retrieved using a combination of six of the MODIS channels between 550 and 2110 nm, which may increase the information content of the MODIS retrieval. With six possible wavelengths e.g. used to minimise the error of the observed to modelled reflectances, there may be more scope for matching with wavelengths more sensitive to larger particles. Another candidate is the particle sphericity assumptions made by both the MODIS and SEVIRI retrievals. MISR is the exception in this regard, since, as was pointed out in Sect. , of the eight particle components in the MISR retrieval, two are non-spherical dust analogues. Moreover, the fact that MISR measurements are made at multiple viewing angles allows for increased resolution of the observed aerosol particle scattering and hence a more constrained knowledge of the phase function. The spherical assumption implies that light scattering occurs due to Mie scattering, which may lead to larger errors at specific scattering angles. Looking again at the AERONET comparisons in Fig. , the points have been colour coded by scattering angle between the Sun and the satellite detectors. Note that the colour coding is different between MODIS and SEVIRI due to the very different scattering angle ranges observed by each. While SEVIRI reveals no apparent pattern in retrieval quality with respect to scattering angle, at high AERONET AODs, the MODIS retrievals show more divergence due to scattering angle, between the high bias of the points at ∼ 100∘ and the low bias of those at ∼ 160∘. Meanwhile, the one clear outlier in the MODIS retrievals with respect to the Abu Al Bukhoosh AERONET data (Fig. b) is also within the 96–110∘ range. Given the limited sample size, it is hard to draw too strong a conclusion from this.

To increase the sample size of the scattering angle range, Fig.  revisualises Fig.  to learn more about the MODIS and SEVIRI scattering angles with respect to high MODIS and SEVIRI AODs. Looking at April, for SEVIRI AOD retrievals in excess of ∼ 0.5, there is a systematic tendency for the co-located MODIS AODs to increase with reducing MODIS scattering angle, which may result in the positive MODIS versus SEVIRI offsets seen in Figs.  and at high AODs. The scattering angle range of 100–120∘ has previously been diagnosed as a potential source of positive offsets when spherical dust assumptions are used by satellite retrievals e.g.. Examples of the differences in the phase functions between spherical and non-spherical dust can be seen in Fig. 4 of and Fig. 2 of , both of which indicate that up to a scattering angle of ∼ 80∘ there is very little difference in the phase functions (at 870 and 550 nm, respectively). The phase functions diverge at higher angles, with the spherical phase functions having the smaller values between ∼ 80–130∘, and the larger values at angles ≳ 150∘. At ∼ 120∘ is the maximum underestimation of the spherical phase function with respect to the non-spherical values, which may lead to a corresponding overestimation of the AOD, especially apparent in Fig. a; simultaneously, the SEVIRI scattering angles are at their highest (Fig. b) in a range where the phase functions may be overestimated and hence the AODs underestimated. This combination of potentially overestimated MODIS AODs and potentially underestimated SEVIRI AODs may contribute to the diversion between the two AOD datasets in this regime.

The AERONET and MODIS retrievals can provide more information about the aerosol content; for example, both retrieve the aerosol Ångström coefficient. For the AERONET comparisons described in Sect. , the mean Ångström coefficients for MODIS and AERONET over KAUST are 0.45 and 0.36, respectively, while over Abu Al Bukhoosh they are 0.51 and 0.35; the greater Ångström coefficients retrieved by MODIS implies that the MODIS retrieval is assuming a smaller aerosol size than AERONET. That the MODIS-retrieved Ångström coefficients are persistently high has also been noted by . The correlations are quite weak at 0.42 and 0.22. Could the possible overestimation of the MODIS Ångström coefficients have an impact on the AOD comparisons, given that the Ångström coefficient is used to scale the MODIS AODs? The MODIS AOD at 550 nm is multiplied by a factor of (550/630)α, where α is the Ångström coefficient. If we consider the mean Ångström coefficients at KAUST and if we use the mean MODIS value of 0.45 then this factor is 0.941; if we use the mean AERONET value of 0.36, then the factor is 0.952. Hence, the MODIS “overestimation” of the Ångström coefficient implies an underestimation of the AOD at 630 nm. The Ångström coefficient does not contribute to the elevated MODIS AODs.

Exploring the aerosol size further, the retrievals can also be subdivided into the aerosol fine and coarse modes using output from the AERONET spectral deconvolution algorithm (SDA; ), and retrievals of fine and coarse-mode MODIS AODs derived from the retrieved fine-mode fraction e.g.. These comparisons are carried out at 500 nm and are listed in Table . Quickly apparent is the strength of the agreement between the coarse-mode retrievals as compared with the fine-mode retrievals, indicative of the dominance of the coarse-mode dust to the AOD over these AERONET sites. Strikingly, the MODIS biases are greater for the fine mode than for the coarse mode despite the fine-mode AERONET AODs being of an order 2–3 times less than the coarse-mode AODs. This MODIS overestimation of the fine-mode AODs with respect to the AERONET SDA algorithm has been noted before by , who observed an overestimation of the MODIS fine-mode fraction of the order of 0.1 for dust-dominated conditions. Similarly, noted that the MODIS AOD retrieval has a tendency to underestimate the size distribution. This may also be related to the higher MODIS Ångström coefficients mentioned earlier. The overestimation in the MODIS fine-mode fraction contributes over half of the bias in the total AODs; in contrast, the MODIS and AERONET coarse-mode AODs are more highly correlated with each other and are a smaller contributor to the overall bias.

A striking feature of Fig.  is the contrast in dust loading between the Red Sea and the Persian Gulf (referred to as “the Gulf” hereafter), both largely enclosed basins on either side of the Arabian Peninsula. The Persian Gulf, also known as the Arabian Gulf, is subject to atmospheric flows and dust storms emanating from Iraq e.g. and Saudi Arabia and occasionally also from Iran. Closer to the active source areas, the Gulf is often affected by Middle Eastern dust storms earlier during their life cycle than is the Red Sea. As with the Red Sea, the Gulf has quiet dust activity during autumn and winter, but there is a high baseline of dust presence throughout the Gulf in spring and summer. This is less intense than in the southern Red Sea in summer, but is a more prolonged feature, and has a more homogeneous spatial pattern within its basin. Due to its closer geographical proximity to the northern Syrian and Iraqi dust sources which flow directly into it, and due to the wind patterns in this region , the Gulf is more heavily influenced by the springtime dust events from the Middle East than the Red Sea.

As presented in Fig. b, dust activity peaked in the Gulf in spring/summer 2009 and in March 2012. The former period was a period of intense dust activity for Iraq and Iran, especially in July 2009: high PM10 (particulate matter less than 10 µm in diameter) concentrations in Tehran were recorded at over 250 µg m-3 during this dusty period . Meanwhile, the latter peak in the Gulf was the month of a large dust event which swept down over the Arabian Peninsula and eventually to the Red Sea, originating in Syria and Iraq. Simulations suggest that this particular event deposited 3.0 Mt of dust to the Arabian Sea, 2.2 Mt to the Gulf, and 1.2 Mt to the Red Sea. This contrast between the Gulf and the Red Sea is broadly supported by the March 2012 monthly mean AODs; from SEVIRI, these are 0.99 over the Gulf and 0.42 over the Red Sea.

How does this contrast affect the satellite retrievals? The SEVIRI/MODIS comparison statistics across the entire Persian Gulf show broad similarity with the Red Sea statistics in Fig. . Over the Gulf, the SEVIRI-MODIS offset at MODIS AODs < 1 is -0.010 and at higher AODs it is -0.378; over the Red Sea, these are -0.010 and -0.416. The individual monthly statistics for both seas are much the same except for a greater prevalence of intense dust events in January over the Gulf than over the Red Sea. For January, April, July, and October, the respective offsets are -0.002, -0.012, -0.040, and -0.0004. What this therefore suggests is that both retrievals' performance over the Gulf are broadly in line with their performance over the Red Sea. This is particularly reassuring for SEVIRI given its geostationary orbit, which is at the limit of its viewing capabilities over the Persian Gulf, with viewing zenith angles > 60∘.

The Abu Al Bukhoosh AERONET site provides additional information to support this argument as a maritime site located far from shore on an oil platform in the southern Persian Gulf (25.50∘ N, 53.15∘ E) with L2 data available in the 2005–2015 period from November 2006 to September 2008. MODIS and SEVIRI validation statistics and scatterplots reveal a similar picture (Fig. b and d) as is present over KAUST, although there are fewer substantial dust events observed. The RMSDs show very minor differences to those seen at KAUST, although it is interesting to note that the biases are somewhat higher: for MODIS, these are +0.08 at KAUST and +0.13 at Abu Al Bukhoosh, while for SEVIRI, these are +0.00 and +0.08. That both retrievals see this increase in bias suggests some common factor influencing the retrievals, perhaps due to local environmental conditions or due to the nature of the aerosol present. The mean MODIS Ångström coefficient is 0.51, as opposed to 0.45 over KAUST. This implies firstly that MODIS is assuming smaller dust particles over Abu Al Bukhoosh than over KAUST, and that the MODIS AOD-scaling process multiplies the 550 nm AOD by a smaller value than over KAUST.

The difference in aerosol type between the two seas becomes more apparent when we compare the MODIS-retrieved Ångström coefficients across the entirety of the two seas. For the 11-year period of 2005–2015, over the Red Sea, the mean Ångström coefficient is 0.66 (0.67 for low AODs < 1, 0.14 for high AODs > 1, as discussed in Sect. ). In contrast, over the Persian Gulf, the mean is 0.96; for low AODs it is 0.99 and for high AODs it is 0.26. For clarity, the basin-scale comparisons are not filtered by Ångström coefficient values < 0.6 as are the AERONET comparisons. The Persian Gulf is clearly a more industrial environment than is the Red Sea, as a centre of global oil extraction both on the surrounding land and within the Gulf itself. As a result, smaller industrial aerosols are a greater contributor to the aerosol loading over the Persian Gulf than over the Red Sea, as evidenced by higher Ångström coefficients, consistent with previous AERONET measurements of the Ångström coefficient from Bahrain, as reported by .

Satellite retrievals of AOD from SEVIRI over the Red Sea show a clear climatological pattern of high summertime atmospheric dust loading over the southern part of the basin, which contrasts with much reduced activity in the north. This pattern induces marked differential radiative heating over the Sea (B15). In this study, we have extended the temporal range of the AOD observations over the Sea to cover the 11 years from 2005 to 2015 and, importantly, evaluated their performance against an expanded set of “ground-truth” measurements. In addition, in light of previously identified biases between different aerosol retrieval algorithms in the presence of dust e.g., we have compared retrievals from MODIS Collection 6 and MISR with the SEVIRI observations.

There is a high degree of inter-annual variability present in this summertime latitudinal gradient of dust presence over the Red Sea, manifested most clearly in the south in July, where band monthly mean AODs range from ∼ 0.5 in 2007 to ∼ 1.8 in 2009. The MACC reanalysis dataset for these two years provides the suggestion that July dust transport over the Sea is dominated by dust sources in the Arabian Peninsula at high-altitude atmospheric layers, whereas near-surface dust tends to be African in origin. Over the past 11 years of the satellite records, some patterns in dust activity can be discerned. The increase in dust activity over the Arabian Peninsula from 2007 to 2013 can be argued to have had an impact on dust loadings over the Red Sea, but unlike over the neighbouring desert regions to the east, this cannot be said to have been a “regime shift”. The summers of 2008–2013 have been particularly dusty over the southern Red Sea, judging by the SEVIRI-retrieved AODs, but recently this has tapered off slightly in 2014 and 2015. Similarly, the Persian Gulf has seen a number of increases in dust activity, such as in 2008–2009 and 2012. This is consistent with the analysis of , who identified a positive dust AOD trend over Saudi Arabia and the wider Middle East from 2000 to 2012, a trend which has been interrupted by reduced AODs over the last few years. However, we also note a recent uptick in dust AODs over both seas in spring 2015.

Over the Red Sea, there is broad agreement between SEVIRI, MODIS, and the surface datasets, although there has been a marginal increase in MODIS AODs by ∼ 0.01–0.02 from the previous Collection 5 dataset (B15). The 11-year satellite retrieval record has enabled the identification of a pronounced positive AOD offset by MODIS with respect to SEVIRI. At lower AODs of < 1, SEVIRI and MODIS are in good agreement, with a MODIS-SEVIRI offset of +0.01. At higher AODs of > 1, the retrieved AODs diverge substantially, leading to inter-retrieval offsets of +0.42: at such high loadings the offsets appear to have a pronounced and systematic optical depth dependence. This may be a consequence of the spherical dust assumptions used by the retrievals, leading to variable sensitivity to dust presence at different sun–dust–satellite scattering angles, an issue which merits further investigation in future studies. Leading on from this, a related open question for future studies is whether this sensitivity is influenced by the origin of the dust, whether from Africa or from the Arabian Peninsula; particle shape and mineralogy may be different between the two source regions. Comparisons with co-located MISR measurements at high AODs indicate that SEVIRI is offset to a value of -0.06 with respect to MISR, while MODIS is offset to a value of +0.19: these are broadly opposite and similar in magnitude, although SEVIRI does show more consistency between AOD regimes. Meanwhile, inter-retrieval comparisons carried out over the Persian Gulf indicate the consistency of the behaviour of the retrievals over a similar enclosed sea affected by dust outbreaks from its neighbouring deserts. Despite the Gulf's proximity to the far edge of SEVIRI's field of view, SEVIRI and MODIS display very similar patterns with respect to each other as they do over the Red Sea, indicating that over both seas the dominant source of any discrepancies between the two retrievals arises simply from the magnitude of the dust activity. Whichever retrieval is most accurate, these differences in behaviour at such high dust loadings will have consequences for the magnitude of the associated differential radiative heating of the surface and atmosphere over the Red Sea and hence the atmospheric and oceanic circulation response.