1 Introduction

Heterogeneous reactions with cloud droplets and aerosols are found to be important for atmospheric chemistry. Since the seminal work of (Dentener et al. 1996), special emphasis has been given to reactions on mineral dust aerosols because these particles provide a large surface area for heterogeneous reactions in the troposphere. Indeed, mineral aerosols produced from windblown soils are an important component of the earth-atmosphere system which is estimated to be between 1000 to 3000 Tg emitted annually (Usher et al. 2003). Convincing evidence that heterogeneous chemistry on mineral dust significantly alters the concentration of important atmospheric gases has been first established by model settings and laboratory experiments. For instance, the heterogeneous chemistry occurring on dust particles acted as a sink reducing ozone (O3) by up to 9 % and sulfur dioxide (SO2) by up to 27 % in an Asian dust storm travelling over the Pacific (Wang et al. 2012). It has also been found that heterogeneous chemistry, including that of SO2 and NOx on dust, plays a significant role in regional haze formation in China (He et al. 2014). However, the magnitude of the uptake simulated by the models is still uncertain because it depends on various environmental conditions and on the uptake coefficient (γ) derived from laboratory experiments. Indeed, the uncertainties in γ are very large as they can cover more than three orders of magnitude for certain species (Zhang and Carmichael 1999; Bian and Zender 2003). For example, some studies have reported values ranging from 2.0 × 10−6 to 2.5 × 10−3 for O3 and from 2.0 × 10−6 to 1.6 × 10−2 for HNO3 (Goodman et al. 2000; Underwood et al. 2001; Michel et al. 2002). In their coated-wall flow tube experiments performed with a variety of Saharan dust samples, Ndour et al. (2008) highlighted the effect of light on the photochemical uptake of NO2: as compared to dark conditions, this uptake could be multiplied by a factor of 8 to 15 depending on the source region. Zhang et al. (2019) also showed that the uptake is greatly enhanced by small amount of surface-adsorbed water because of the exponential relationship between SO2 and Relative Humidity (RH). Finally, there is also evidence of a synergistic effect when various trace gas species coexist like NOx with SO2 (He et al. 2014) or NO2 and ammonia (NH3) (Zhang et al. 2019). All these laboratory studies suggest that dust storms could have the potential to change the distribution of reactive trace gases in the troposphere but there is still a need to evaluate the real-life impact of natural dust on atmospheric processes.

Xie et al. (2005) explored at three sampling sites in Beijing (China) the changes in reactive trace gases before, during, and after a dust storm. To our knowledge, no other urban scale study has evaluated those effects from an in-situ field perspective especially in the megacities of Africa and the Middle East where mineral dust is present and where anthropogenic emissions are expected to increase dramatically in the next century (Thera et al. 2022; Salameh et al. 2017; Knippertz et al. 2015). Greater Cairo (GC), is a good example of these megacities. It consists of three different districts (Cairo, Giza, and Kalubia), its population approaches 22 million and hosts more than one- third of the total industrial activity of Egypt (Hereher et al. 2021). As a consequence, GC suffers from high mass concentration of atmospheric pollutants such as airborne particulate matter (PM), carbon monoxide (CO), nitrogen oxides (NOx), ozone (O3) , sulfur dioxide (SO2) and organic compounds (OC) (Abu- Allaban et al. 2002, 2009; Hassan 2018a, b, c; Marchetti et al. 2019; Shaltout et al. 2020; El-Dars et al. 2004; Talaat et al. 2021). The origin of PM is both natural (desert dust) and anthropogenic (traffic, industry, biomass and waste burning) (Abu-Allaban et al. 2007; El-Metwally et al. 2011; Favez et al. 2008a, b; Mahmoud et al. 2008; Shaltout et al. 2018a, 2019). In their recent analysis of the spatial and temporal variability of the PM10, SO2, NO2, CO and O3 mass concentrations at the 17 sites of this study, Mostafa et al. (2019) suggested that biomass burning (sulfur-containing fuels) by industry and transportation sectors is the main source of SO2 and NO2. Being located downwind of the Saharan Desert, GC also experiences regularly high dust loadings (Hassan and Khoder 2017; Eltahan et al. 2018; Hassan 2018b; Shaltout et al. 2018b), which makes it a perfect place for studying the potential interactions between pollution and mineral dust.

In this work, we propose to use a statistical approach based on in-situ observations to detect any effect of reactive trace gas (SO2, NO2, and ozone) uptake by mineral dust at an urban scale. A multi-pollutant and multi-year database of gaseous (NO2, SO2, O3) and particulate (PM10) concentrations measured in the Greater Cairo area has been collected and quality-checked in combination with the meteorological measurements of the Egyptian Meteorological Authority (EMA). The objective was to analyze the spatial and temporal variability of trace gases during dust events with the underlying question: are we able to evidence any impact (surface uptake) of the presence of dust on the ozone, NO2, and SO2 concentrations?

2 Data and methods

The methodology consists of i) a quality check and selection of the air quality data between 2010 and 2014, ii) extracting representative stations by agglomerative hierarchical clustering, ii) isolating dust events on the basis of PM10 in-situ measurements, remote sensing (sunphotometer and satellite) observations, and air mass back-trajectory analysis, and iv) applying a non-parametric hypothesis test (the Kruskal Wallis – KW test) to compare trace gas concentrations during dust and non-dust events. In the following we will first describe the data before detailing the methodology.

2.1 In-situ observations and clustering of the ground stations

The in-situ air quality observations are performed at the ground stations of the Egyptian Environmental Affairs Agency (EEAA) and Egyptian Meteorological Authority (EMA). The Air Quality Network data include the hourly mass concentrations of ambient sulfur dioxide (SO2), nitrogen dioxide (NO2), ozone (O3), and PM10. In this study, we use the concentrations of 16 Greater Cairo sites of the EEAA and one of the EMA (Abbasseya) in the period 2010 to 2014. Figure 1 and Table 1S in the supplementary material report all information about these monitoring stations.

Fig. 1
figure 1

Map of the 17 stations in Greater Cairo (left) and better resolution map indicating the location of each station within the greater Cairo region (right). The color code indicates the typology of the stations according to the classification of the EEAA and EMA

Full size image
Table 1 Average concentration of ambient air pollutants during the selected dust cases at the stations representative of the 3 PM10 clusters identified by the hierarchical clustering
Full size table

The recovery rate of the measurements for each year and each station have been evaluated. In terms of number of stations and available data, 2012 shows the highest recovery rate (77% on average). Only the 6th October station had no validated SO2 and NO2 data in this year, whereas data from 3 to 4 stations were missing in the other years. Therefore, 2012 was selected for the following analysis. The recovery rates and the number of values exceeding at each site the daily SO2, NO2 and PM10 limits of the air World Health Organization quality guidelines (WHO 2021, https://www.who.int/news-room/fact-sheets/detail/ambient-(outdoor)-air-quality-and-health) are reported in Table 2S.

Table 2 Kruskal-Wallis results for comparing weighted SO2 concentrations during non-dust days and after adding a dust day to the test at the 4 sites representative of each cluster
Full size table

At most stations, the levels of PM10 always exceed the daily guideline value (45 µg.m-3), thus confirming the very high PM10 loading in Greater Cairo already highlighted in previous studies (Favez et al. 2008a, b; Kanakidou et al. 2011; Wheida et al. 2018; Zakey et al. 2008).

Except at the Masara station, from 3% to 28% of the SO2 recorded data are higher than recommended by the WHO values (40 µg.m-3) but the level of exceedance is variable from one station to the other and depends on their typology. For NO2, the levels remain usually higher than the WHO threshold (25 µg.m-3) with exceedances ranging from 46% to 99% except at the Shoubra El-Kheima station where only 6% of the recorded data are higher than this limit.

In order to reduce the number of stations retained for the analysis of the effect of dust on trace gas concentrations, we used Agglomerative hierarchical clustering (hereinafter referred to as AHC) to classify the 17 sites into homogeneous groups with similar features. Indeed, AHC is a well-established associativity analysis methodology allowing the identification of the inherent or natural groupings of objects (Johnson and Wicherrn 2007). This multivariate method has come to be recognized as an effective statistical tool to deal with tasks as grouping time series (e.g., Gramsch et al. 2006; Afif et al. 2008; Govender and Sivakumar 2020). In AHC, each pollutant is represented by a matrix of up to 17 columns (one for each station with data in 2012) and a number of rows corresponding to the number of hourly measurements available in the year 2012. For instance, 1589 PM10 measurements at 17 sites, 1696 SO2 measurements at 15 sites, and 4754 NO2 measurements at 13 sites are available in 2012., we used the XLSTAT 2016 software from AddinSoft to calculate dissimilarities of squared Euclidean distances then Agglomerative Hierarchical Clustering according to Ward (1963).

2.2 Identification and characterization of the dust events

The massive advection of mineral dust produced by wind erosion in the deserts located upwind of Cairo leads to a sharp increase (up to several hundreds of µg.m-3) of the PM10 concentrations recorded at the GC air quality monitoring stations. These events are generally caused by eastbound depressions and are associated with strong west/southwest winds and dry air masses (El‐Askary and Kafatos 2008; Eltahan et al. 2018). Therefore, the detection of dust cases will be based primarily on the measurement of unusually large PM10 concentrations (i.e., >95th percentile of the yearly measurements) coinciding with back trajectories confirming that the air masses had been travelling above desert areas before reaching GC. These air-mass back trajectories are calculated and plotted with the HYSPLIT model (Stein et al. 2015) using the ZeFir tool under Igor Program (Petit et al. 2017).

In a second step, the observations of the Moderate Resolution Imaging Spectroradiometer (MODIS) aboard the Aqua and Terra spacecraft will be used to confirm the presence of plumes of mineral dust above GC. These images were downloaded from the NASA website (https://worldview.earthdata.nasa.gov/).

Finally, the inversion of the measurements of the Aeronet Robotics Network (AERONET) sunphotometer (Holben et al. 1998) operated at the Abbasseya station of the EMA allowed characterizing of the aerosol present in the atmospheric column above GC. In this work, we used the quality-assured (level2, version 3) inversion products downloaded from the Aeronet portal (http://aeronet.gsfc.nasa.gov/), and more precisely the Aerosol optical depth (AOD) at wavelength 500 nm, the Angstrom exponent (AE) calculated between 440 and 876 nm, and the Single scattering albedo (SSA) at 440 nm. Although the sunphotometer measurements are not performed in cloudy conditions or at night, which means that its data are not acquired as regularly as those of the air quality network, these products will be precious as inputs to evaluate the photolysis rate (JNO2) of NO2 with the Tropospheric Ultraviolet and Visible (TUV) radiation model.

2.3 Meteorological conditions

Hourly averages of various meteorological parameters are recorded at the Abbasseya meteorological station located in the core of GC. The seasonal variability of the dry air temperature (T) and relative humidity (RH) in 2012 does not differ from that of the other years of the 2010 -2014 period (Fig. 1S). T varies between 10 and 40 °C in the transitional seasons but is less variable in winter (9-25 C˚) and summer (21-42 C˚). On average, RH is particularly low as it does not exceed 60% in all seasons. The prevailing winds have speeds between 6-9 m.s-1 and are from the north/north-east sector (0-45 degrees) (Fig. 2S). However, particularly strong (speed 9-12 m.s-1) south-western winds were recorded in winter (Fig. 2S).

2.4 Comparison of dusty and non-dusty day concentrations

For detecting any potential impact of mineral dust on the SO2, NO2, and O3 concentrations, we compared their concentrations during dust events with those of non-dust periods. However, this comparison only makes sense if the effect of confounding factors has been accounted for. Indeed, the variability of trace gas concentrations is driven by numerous factors such as their direct primary emissions, chemistry, and dilution. Therefore, to detect any potential effect of dust surface on the uptake of reactive trace gas, one should make sure that the comparison with non-dust cases is performed in similar conditions. In the following, we first list the factors that need to be controlled, then present the statistical tests used for the comparison of the dust and non-dust cases. The evaluation of the role of the potential confounding factors will be discussed in the Results section.

2.4.1 Major confounding factors

In urban areas, the emission of primary pollutants or of the precursors of secondary ones usually varies with the hour of the day, the day of the week (weekends or holidays versus business-as-usual days), or the season. In order to minimize the bias introduced by differences in the emissions, we compared the gas concentrations of dust days with those of the same days and hours of the previous and following weeks. In doing so, we implicitly assume that for a given day and hour the emissions do not vary considerably from one week to the other, and thus avoid the risk of comparing a working day with a none working one, which could have occurred if we had compared the dust day with the previous and the following days.

Meteorology is also expected to affect trace gas variability. Wind speed (U) and the height (H) of the planetary boundary layer (PBL) play an obvious role in the concentrations of the gaseous species because they control their dispersion. The combined effects of U and H are often quantified by the means of the so-called ‘Ventilation Index’, which is simply equal to their product (Mahmoud et al. 2008). Because the uptake of SO2 by mineral particles is enhanced by the presence of water on their surface, the influence of potential differences in atmospheric relative humidity between dusty and non-dusty days must also be evaluated. The basic meteorological variables necessary for these quantifications are measured at the EMA’s site of Abbasseya (relative humidity, wind direction…) or available (U, H) as products of the ERA5 reanalysis of the European Centre for Medium-Range Weather Forecast (ECMWF). These products were downloaded from https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-single-levels?tab=form.

Finally, by absorbing and scattering solar radiation the presence of the dust aerosol layer could alter the photolysis frequency of one of the target trace gas, namely NO2 (JNO2). Indeed, during a Saharan dust event of the MINATROC campaign in Cape Verde a reduction of 17% was observed for J(NO2) (De Reus et al. 2005). Therefore, the potential effect of dust on JNO2 needs to be evaluated. Due to the absence of radiometric J(NO2) measurements, this effect was quantified by running the on-line TUV model version 5.3 available at https://www.acom.ucar.edu/Models/TUV/Interactive_TUV/. Besides date and time, the input data are latitude, longitude, altitude, surface albedo, the wavelength range of calculation (300-400 nm), and the optical properties of the aerosols (AOD_500 nm, AE_440-870 nm, and SSA_440 nm). Those parameters have been specified for EMA’s site of Abbasseya where the AERONET sunphotometer is located.

2.4.2 Statistical comparison of dust days versus non-dust days

The comparison of trace gas concentrations (corrected for the effect of confounding factors as detailed in results section during dust and non-dust days was performed for the stations representative of the clusters identified by the AHC. In order to choose the most appropriate statistical test for this comparison, we first evaluated the normality and lognormality of the distributions by the Shapiro-Wilk test (Shapiro and Wilk 1965). Because the test revealed the absence of normality or lognormality in these distributions, we decided to apply the test of Kruskal-Wallis (KW) (Kruskal and Wallis 1952). Indeed, being a nonparametric method, the Kruskal–Wallis test does not assume a normal distribution of the residuals and can be used to test the equality of medians of a group of independent samples. The null hypothesis is that the medians of the tested group are equal. The alternative hypothesis is obtained when at least one sample is significantly different from the others.

Before performing the comparison with the dust days, the equivalence of the concentrations of the non-dust days used as the reference was checked by a first application of the KW test. The significance was assessed at the 95 % confidence interval: then, if the null hypothesis was confirmed, (p > 0.05), the test was applied in a second step to the whole group of non-dust and dust days to check if the dust day was different from the others.

3 Results and discussions

3.1 Agglomerative hierarchical clustering

3.1.1 PM10

The results of the application of AHC to PM10 are reported on Fig. 2. The dendrogram (upper panel) shows that the 17 stations are distributed into three clusters of different importance: class 1 (C1, 9 stations), class 2 (C2, 7 stations) and class 3 (C3, only the Qulaly station). The dotted line represents the truncation which led to three homogeneous classes. The stations most representative (the so-called ‘central objects’) of C1, C2, and C3, are El-Salam, Masara, and Qulaly, respectively. The dissimilarity between C1 and C3 is larger than the one between C2 and C3, which suggests that the levels and the variability of PM10 concentrations at Qulaly station resemble more those of the 7 stations of C2 (and particularly Abo Zaabal, which is very close to Qulaly on the dissimilarity scale) than those of the 9 stations of C1.

This clustering is consistent with the different ranges of PM10 levels observed at the stations and their three distinct types of diurnal variability (lower panel of Fig. 2). In C1, PM10 is the lowest and varies between 73 and 261 µg.m-3. These -relatively- moderate levels and their lack of clear diurnal variability (Fig. 2) suggest that the C1 stations are not under the direct influence of any specific source of PM and that they can be considered as being representative of the PM10 background levels in the GC area. Noteworthy, this could be consistent with the fact that the 9 stations of C1 are not located in the very heart of the GC domain. Three of them are even at its western (6-Oct station) or eastern (El-Salam and New Cairo stations) boundaries. On average, the levels of PM10 mass concentrations in C2 (172 to 404 µg.m-3) are higher than in C1. Of the 7 sites of C2, 4 are located near industrial facilities: namely, Abo Zaabal, Shoubra El-Kheima, Helwan, and El-Tbeen. The 7 sites are distributed along a north (Abo Zaabal) to south (El-Tbeen) axis. As highlighted by (Mostafa et al. 2019), the higher levels of PM10 along this axis can be explained in part by the neighboring to near industrial regions but also by the dominant northern wind (Fig. 2S). The diurnal variations of PM10 in C2 display two peaks: a well-marked morning peak at 10 AM and an evening peak at 8 PM separated by a drop in the middle of the day. This typical diurnal variability is related to variations in both the anthropogenic activities and boundary layer height.

Fig. 2
figure 2

a Dendrogram for PM10 data (upper panel, in which the dotted line represents the truncation level) and b average diurnal variability of PM10 in each cluster and at representative stations (lower panel). Central object represents the object nearest to the centroid in each class

Full size image

The Qulaly monitoring station is the only one representing C3. Its PM10 diurnal variations are similar to those of C2, but it is distinguished by the AHC because its hourly average levels are considerably larger (268 to 449 µg.m-3) than in C2 and its mid-afternoon drop is less pronounced.

3.1.2 SO2

The results of AHC on SO2 data are shown in Fig. 3. The dendrogram shows that the 15 stations are distributed in four clusters. Two stations are isolated into one single cluster: Qulaly (C4) as for PM10 and Abo Zabaal on the North-Eastern border (C2). Both stations show the highest SO2 levels (19-33 µg.m-3 and 17-36 µg.m-3, respectively) but their diurnal variabilities differ. Abou Zabaal displays a bimodal distribution with two well-pronounced morning and evening peaks similar to those of PM10 in cluster C2 (Fig. 3), whereas no mid-afternoon drop is observed at Qulaly. C1 (8-28 µg.m-3) and C3 (13-32 µg.m-3) contain 7 and 6 stations, respectively. The clustering results cannot be explained by the typology or geographical location of the sites but what is observed is that the variance between stations in C1 is lower than the variance between stations grouped in C3 (Fig. 3).

As for PM10, C1 shows a background-like variability with only a small increase of SO2 between 06h and 10h local time (morning rush hours). At all other stations, the morning peak is more pronounced. In C2, and to a lesser extent in C3, a second increase of SO2 is observed between 18h LT and 20h LT. As described by Mostafa et al. (2019), the combustion of sulfur-containing fuels (e.g., oil and diesel) might explain the morning and late evening peak concentrations. Those peaks are enhanced at traffic sites like Qulaly. As for PM10, the drop in the middle of the day is attributed to stronger vertical mixing.

Fig. 3
figure 3

a Dendrogram for SO2 data (upper panel, in which the dotted line represents the truncation level), and b average diurnal variability of SO2 in each cluster and at representative stations (lower panels)

Full size image

3.1.3 NO2

The results of AHC on NO2 data are reported in Fig. 4. Most of the sites (9 out of 13) were grouped into one class (C1), C3 contains 2 stations, C2 corresponds to Abo Zaabal, and C4 to Nasr city where 92% of the data were greater than the WHO hourly limit (Table 2S). The C1 stations show a background-like variability characterized by a smooth increase during the course of the day with a maximum of around 18h LT indicating the secondary production of NO2. The third cluster (C3) contains Qulaly and Abbasseya where the diurnal variability was the same during night hours and similar during the day but the levels at Qulaly were higher. All sites show a bimodal variability at traffic rush hours (06h to 10h LT and 16h to 20h LT). However, at Nasr City (C4) and Abo Zabaal (C2), the night-time levels of NO2 are larger than those observed in the afternoon.

Fig. 4
figure 4

a Dendrogram for NO2 data (upper panel, in which the dotted line represents the truncation level), and b average diurnal variability of NO2 in each cluster and at representative stations (lower panel)

Full size image

In summary, on average and for the year 2012, most of the stations either show a background like or a bimodal variability with concentrations peaking at traffic rush hours. This behavior is observed also for the secondary product NO2. These commonalities can most probably be explained in part by traffic emissions, which are one driver of the diurnal variability, but the role played by the boundary layer dynamics is also crucial as will be shown below.

3.2 Identification of dust events from in-situ and remote sensing observations

As explained in the Methods section, dust events are primarily extracted when PM10 exceeds the 95th percentile (446 µg.m-3) of the measurements performed at all stations. The analysis of air mass trajectories with the ZeFir tool (Petit et al. 2017) based on the HYSPLIT model and the examination of MODIS images are also used to support the dust event selection (Figs. 3S and 4S). The application of this procedure yields seven dust events distributed over eight days (or dust cases) whose details are reported in Table 1 for the three PM10 Central Object stations (El-Salam, Masara and Qulaly) extracted by the AHC. Consistent with Eltahan et al. (2018) who noted that dust storms usually occur in winter or spring in GC, four of the 2012 events occurred in wintertime and three in springtime. The air-mass back trajectories and the MODIS images confirm the Saharan origin of dust that came either from the south-western desert of Egypt or from the Libyan Desert.

Wintertime dust events were characterized by a larger RH (38-60%) than the springtime (23-37%) ones. Overall, RH is usually lower than 60% and close to 40% on average. The duration of the events ranged from three hours (case 7, at night) to more than one day (case 6, which is therefore split into 6A and 6B in Table 1). In all cases, the largest average PM10 are recorded at Qulaly. The average PM10 at El-Salam compares to that of Masara. The range of PM10 variations is from 228 ± 236 µg.m-3 at El-Salam (case 1) to 1540 ± 562 µg.m-3at Qulaly (case 6-A). The available levels of SO2, NO2 and O3 are also reported. During the dust events, the wind speed (WS) can reach 12 m.s-1 on average (cases 3, 4 and 5). The largest values of the ventilation index coincide with these strong wind conditions.

Figure 5 illustrates the temporal evolution of the atmospheric conditions before, during, and after the 6-A and 6-B dust cases at a selection of stations including the PM10 clustering central object stations. During the two dust events, all stations show a consistent and remarkable tenfold increase in their PM10 loadings. The increase of PM10 is correlated with the increase of the wind speed recorded at the Abbasseya station (up to 10 m.s-1) and the wind direction shifted to the Western sector a few hours before, pointing out the Saharan desert as the source of dust. The wind direction observed in situ is consistent with the trajectory analysis in the Supplement Material (Fig. 4S). During case 6-B, the AOD and AE values are available and equal to 0.8 and 0.0, respectively. The very low value of the AE, characteristic of coarse particles, confirms the dust-like nature of the aerosol. Outside the dust event, the AOD and AE are below 0.5 and above 0.5, respectively, indicating a rather mixed aerosol (El Metwally et al. 2008).

Fig. 5
figure 5

Example of time series of PM10, wind speed (WS), wind direction (WD), AOD and Ångström coefficient for the Cairo stations representative of the three clusters before, during and after the dust events of 18 and 19 April 2012 (cases 6-A and 6-B in Table 1). The dust event is color-coded in orange

Full size image

3.3 Evaluating the diurnal variability of trace gas concentrations

A common and simple approach like the one described in Xie et al. (2005) consists in comparing the absolute concentrations of reactive trace gases before, during and after the dust event in order to test whether a decrease in their concentrations is observed when dust is present. Our purpose here is to illustrate how much this approach is challenging regarding the high temporal (hourly) variability of the reactive trace gases close to anthropogenic emission sources. An example of the hourly variability is illustrated in Fig. 6 for dust event n°5 at the Abbasseya station. The diurnal variability of the hourly SO2, NO2 and ozone concentrations for the non-dust days of April 2012 is compared to that of the dust day (April 1st). Note that to facilitate the comparison of days with different levels of SO2 concentrations, the SO2 data of Fig. 6 have been normalized to their average night-time values.

Fig. 6
figure 6

Diurnal variations of the three trace gases, PM10, wind speed and ventilation index at Abbasseya station in April 2012 during non-dust days and during the dust storm of 1 April 2012 (case 5). The hourly concentrations of SO2 are relative to the night-time values

Full size image

The daytime levels of all the trace gases during the dust event of April 1st (dotted lines) tend to be lower than during the non-dust days (solid lines). However, they remain within the limit of one standard deviation (shaded areas), which is quite large. The drop in ozone concentrations is coincident with the sharp increases in the wind speed, ventilation index and PM10 levels, but the decrease of the NO2 and SO2 concentrations precedes them by 2 to 4 hours, respectively. Therefore, the drop in concentrations cannot be explained only by the ventilation effect or the increase of the PM10 concentration. This example illustrates the complex interplay between the various source- and sink-terms governing the diurnal variability. In a nutshell, the important day-to-day and hourly variability of the trace gas concentrations in urban areas makes it difficult to ascribe any clear drop of these concentrations to the advection of dust. In particular, one cannot compare absolute concentrations during the few hours preceding and following the dust storms, as in Xie et al. (2005). In order to minimize the effect of these day-to-day and hourly variations of anthropogenic activities, we assumed that their pattern did not change significantly from one week-day to the same week-days of the same month and therefore decided to compare the dust days with the same non-dust weekdays of the preceding and following weeks.

3.4 Evaluating the potential effect of confounding factors

During an intense dust event, the reduction of the photolysis frequency of NO2 [J(NO2)] due to the absorption and scattering of solar radiation by the dense aerosol layer could also play a role. This role was evaluated for the only dust case (6-B) for which Aeronet aerosol optical properties were available. The comparison with non-dust days (Table 3S) suggests that there should be only a minimal (between 1 and 8%) reduction of the photolysis frequency.

Table 3 Kruskal-Wallis results for comparing normalized NO2 concentrations during non-dust days and after adding a dust day to the test at the selected sites
Full size table

Finally, wind speed tends to be larger during dust events than on non-dust days and this favors the dispersion of the trace gases. The variations of the boundary layer height also have direct implications for their concentrations. In summary, large U associated with large H should lead to lower concentrations of SO2 and NO2 than usual conditions. This effect of the ventilation is clearly illustrated by the fact that the largest SO2 (Fig. 7) and NO2 (Fig. 5S) concentrations are systematically associated with low ventilation indices. This emphasizes the need to compensate for the role of UH differences in the comparison of the dust days (usually characterized by large UH) and non-dust ones. In the following, we will take this effect into account by weighing (multiplying) the NO2 and SO2 concentrations by UH.

Fig. 7
figure 7

Scatter plot of SO2 concentrations vs Ventilation index at the 4 representative stations yielded by the AHC. The color code is the same as on Fig. 3 and the filled (empty) markers correspond to dust (non-dust) periods

Full size image

3.5 Comparison of trace gas mixing ratios: dust days versus non-dust days

In this section, we explore the effect on the concentrations of reactive trace gases of the high PM10 loadings observed in Greater Cairo during dust events. As explained above, these concentrations are weighted by UH to account for the impact of possible differences of ventilation index between the dusty and non-dust days. The comparison between dusty and non-dusty days is performed when the Kruskal-Wallis test previously applied to the group of non-dust days used as a reference confirms their consistency of behavior (p > 0.05).

3.5.1 Sulfur dioxide

The results of the statistical tests for the stations representative of the four clusters are presented in Table 2. For the eight non-dust groups and the four stations (32 combinations), 15 have p values larger than 0.05 (5, 4, 3, and 3 cases at Nasser Institute, Abo Zabaal, El-El-Salam, and Qulaly stations, respectively). This is an indication that there was no statistical difference between the non-dust days within these groups. On the opposite, the remaining cases (i.e., those for which p < 0.05) show statistical differences that reveal their absence of homogeneity. Those cases were therefore not used for the following comparison with the dust days.

After including the dust days in the 15 tested groups, p remains larger than 0.05 in 8 cases, thus showing that the dust days are not different from the non-dust ones. In the 7 other cases (Table 2), the weighted concentrations of SO2 during these dust events are statistically different from those of the non-dust days but these differences can be due either to an enhancement (5 out of 7) or to a decrease (2 cases) of the weighted SO2 concentrations during the dust event. Figure 8 provides an example of these contrasting behaviors for the Abo Zaabal station. During event 6B, the absolute SO2 concentrations (red points) were lower than during the comparable non-dust week-days and they remain so (blue points) after accounting for the effect of ventilation. In the case of event 3, the SO2 concentrations were similar during the dust and non-dust days, but they become larger after the weighing by UH.

Fig. 8
figure 8

Examples of comparison between non-dust days (ND) and dust day (DD) at Abo-Zabaal station. Error bars are the standard deviations of SO2

Full size image

This suggests that the decrease of the SO2 concentrations observed during some dust events in Cairo is rather due to the increase of the ventilation index (5 cases out of 7) than to uptake at the surface of the dust particles.

Event 6B (Table 2) is particularly interesting in the sense that it allows comparing the potential impact of mineral dust on the SO2 concentrations at the 4 stations during the same event. After the inclusion of the dust day, the p values yielded by the KW test indicate the absence of effect on the SO2 concentrations at Nasser Institute and Qulaly (p>0.05), but a significant one at (a decrease) at Abo Zaabal and El El-Salam.

3.5.2 Nitrogen dioxide

The same procedure as for SO2 has been applied to NO2 and the results of the statistical tests are presented in Table 3. For the 8 groups of non-dust days at the 4 stations, p-values are found to be larger than 0.05 in 15 cases (3, 2, 5, and 5 cases for New Cairo, Abo Zaabal, Qulaly, and Nasr City, respectively).

After inclusion of the dust days in the 15 tested groups, the resulting p values remain larger than 0.05 in 9 cases. In the other 6 cases (Table 3), p value is < 0.05 in 1, 1, 3, and 1 occurrence at the four stations, respectively. A reduction (by 48%) of the UH-weighted NO2 is only observed at the Qulaly station during case 7 (Fig. 9). In the 5 remaining cases, the difference with the non-dust days is due to an enhancement of the NO2 concentration and not to a reduction.

Fig. 9
figure 9

Examples of comparison between non-dust days (ND) and dust day (DD) at Qulaly station. Error bars are the standard deviations of NO2

Full size image

3.5.3 Ozone

The statistical tests could be applied for three of the dust events and only at Abbasseya, which is the only station where the ozone concentration was measured. The non-dust day p-value is larger than 0.05 only during events 5 and 6B. After inclusion of the dust day in this group, the p value remains larger than 0.05, indicating that there is no significant effect of the presence of mineral dust on the ozone concentration corrected for the effect of the ventilation index (Table 4).

Table 4 Kruskal-Wallis results for comparing normalized O3 concentrations during non-dust days and after adding a dust day to the test at Abbasseya site
Full size table

3.6 Discussion

In 2012, eight dust cases were identified and for SO2, as well as for NO2, the 17 stations of the air quality monitoring network can be sorted into four representative clusters. Therefore, there is theoretically a maximum of 32 (4 by 8) combinations to test the impact of the presence of dust on the trace gas concentrations by comparison with non-dusty days. Practically, after accounting for the missing data and eliminating the situations for which the behavior of the group of non-dust days was not homogenous enough to serve as a sure-proof reference, this number was reduced to 15 for both SO2 and NO2. In slightly more than 50% of the situations (8 and 9 cases, respectively), there was no statistically significant difference in UH-weighted concentrations between the dust and non-dust periods. In the other cases, the majority (5/7 for SO2 and 5/6 for NO2) of the observed differences were not due to a reduction of the trace gas concentrations but to an increase. For ozone, the number of usable test-cases is quite limited (only 2), but in both cases, no modification of the UH-corrected concentration is observed in the presence of dust. These results obtained in natural conditions during intense dust events seem to contradict the current assumption mostly based on laboratory observations that heterogeneous reactions at the surface of dust particles act as a significant sink for reactive trace gases.

One first possible explanation for this apparent contradiction lies in the low atmospheric relative humidity prevailing during the dust events in Greater Cairo. Zhang et al. (2019) showed that the uptake of SO2 on dust particles was facilitated by the presence of NO2 and NH3 and increase of RH above a threshold of about 50% that was barely reached in Cairo during the dust events of this study. In such conditions of dryness, reactions in the aqueous phase are not possible or at least too limited to induce a detectable reduction of the concentrations of the trace gases. However, a systematic study involving more dust cases and, if possible, a larger range of RH conditions would be necessary to confirm this hypothesis.

4 Summary and conclusion

Greater Cairo is a natural laboratory for studying the interactions between reactive trace gases of anthropic origins and mineral dust produced by wind erosion of the deserts surrounding the megacity. In this work, eightintense dust events of 2012 have been identified and selected as case studies to evaluate the impact of the presence of dust on the concentrations of SO2, NO2, and ozone measured at a selection of stations (four for SO2 and NO2, one for O3) representative of the Greater Cairo area. As compared to those of the preceding and following non-dust periods of similar anthropogenic activities, these concentrations are significantly lower during the dust periods. However, our analysis of the role of potential confounding factors (wind speed, boundary layer height, relative humidity, modification of the actinic flux) showed that this drop of the concentrations was a consequence of the larger wind speed and higher boundary layer height prevailing during the dust events and favouring the dispersion of the trace gases. Once this effect is corrected, the application of the non-parametric Kruskal-Wallis test no longer reveals any difference between the dust periods and non-dust reference ones in 8/15, 9/15, and 2/2 cases for SO2, NO2, and O3, respectively. When a statistically significant difference is found, which is due to a decrease of the SO2 or NO2 concentrations in only 2/15 and 1/15 cases and to an increase in 5/15 cases (for both gases). These results show that, in Greater Cairo, there is no detectable effect of adsorption of the reactive gases on the surface of the air-suspended dust particles which are contradictory to laboratory experiments indicating this adsorption. They can be primarily interpreted by the relatively low RH (<60%) prevailing in Greater Cairo, particularly during the episodes of Saharan dust advection, the uptake coefficients of the trace gases are expected to be very low.