PM2.5 surface concentrations in southern West African urban areas based on sun photometer and satellite observations

Southern West Africa (sWA) is influenced by large amounts of aerosol particles of both anthropogenic and natural origins. Anthropogenic aerosol emissions are expected to increase in the future due to the economical growth of African megacities. In this paper, we investigate the aerosol optical depth (AOD) in the coastal area of the Gulf of Guinea using sun photometer and MODIS satellite observations. We use a lightweight handheld sun photometer measuring the solar irradiance at 465, 540 and 619 nm operated manually every day from December 2014 to April 2017 at 5 different locations in Côte d’Ivoire 5 and Bénin. Handheld sun photometer observations are complemented by available AERONET sun photometer observations and MODIS level 3 time series between 2003 and 2018. MODIS daily level 3 AOD agrees well with sun photometer observations in Abdidjan and Cotonou (correlation coefficient R=0.89 and RMSE=0.19). A classification based on the Angstrom Exponent is used to separate the influence of coarse mineral dust and urban-like aerosols. The AOD seasonal pattern is similar for all the sites and is clearly influenced by the mineral dust advection from December to May. AODs are analyzed in coincidence with 10 surface PM2.5 concentrations to infer trends in the particulate pollution levels over conurbation of Abidjan (Côte d’Ivoire) and Cotonou (Bénin). PM2.5 to AOD conversion factors are evaluated as a function of the season and the aerosol type identified in the AE classification. Highest PM2.5 concentrations (up to 300 μg/m) are associated to the advection of mineral dust in the heart of the dry season (December-February). From December to March the median concentration above Abidjan and Cotonou is around 40 μg/m, while it is around 20 μg/m during the rest of the year. Considering only the days during which the AOD 15 belongs to the urban-like aerosol category, we observe a significant trend S=0.32 μg/m/year in the PM2.5 concentrations over the period 2003-2017. This trend leads to an increase of 5 ± 3 μg/m over 15 years and is coherent with the expected increase in combustion aerosol emissions in sWA. 1 https://doi.org/10.5194/acp-2020-617 Preprint. Discussion started: 11 August 2020 c © Author(s) 2020. CC BY 4.0 License.


Sun photometers
summarizes the location, type of instrument and observation periods. We have used different types of sun photometers, 80 automatic and handhelds. The automatic CIMEL sun photometer is the reference instrument used in the AERONET network (Holben et al., 1998) for measuring the AOD and retrieve columnar aerosol optical properties and size distribution. We have used the level 2 quality assured data processed with the version 3 of the aerosol optical depth algorithm (Giles et al., 2019). We used the data for Ghana (station named Koforidua_ANUC located at 6 • 6' N, 0 • 6' W), Nigeria (station named Ilorin located Handheld sun photometer is a well-known scientific instrumentation for measuring atmospheric transmission (Porter et al., 2001;Volz, 1959Volz, , 1974 CALITOO (Djossou et al., 2018). CALITOO operating wavelengths for the CALITOO are 465 nm, 540 nm and 619 nm. The sun photometer measures the Sun irradiance at the 3 wavelengths. The atmospheric optical depth is then retrieved following the Beer-Lambert law knowing the calibration constant for each instrument and the relative air mass. The AOD is then retrieved after subtracting the Rayleigh and trace gases optical depth.
For the HHC, observations were acquired twice a day at around 9:00 and 15:00 UTC. For the CALITOO sun photometer, 95 the observations were acquired at around 13:00 LT. The operators were asked to make measurements only when the sun was not obscured by clouds and have proceed with a sequence of 5 measurements within about 15 minutes. The presence of subvisible cirrus or broken clouds within the field of view induces spurious variation in the atmospheric transmission (Smirnov et al., 2000) that can be easily detected by looking at the standard deviation of the 15-minute series of AOD measurements.
An arbitrary threshold of 0.2 on the standard deviation has been selected to remove the cloud-contaminated observations.

100
The diurnal variability range is expected to be less than 10% for our site conditions (Smirnov, 2002). The sun photometer observations are reported as daily averages.
The total uncertainty in AOD for the AERONET instruments is ±0.01 for λ > 440nm and ±0.02 for shorter wavelengths (Holben et al., 1998). CALITOO sun photometers were calibrated prior to the site deployment using the Langley-plot method (Soufflet et al., 1992;Schmid and Wehrli, 1995) at the Izaña high-altitude observatory (Basart et al., 2009). A direct comparison 105 with a AERONET instrument indicates that the total uncertainty in AOD for CALITOO is ±0.02 for all the wavelengths. The post-field calibration was done using a reference AERONET instrument and indicates a change of about 1% per year in the calibration.
AOD measurements are all reported at 550 nm because this wavelength is a reference for visibility calculation (Boers et al., 2015) and satellite mission (e.g. Remer et al., 2008). The Angström exponent (AE) (Angström, 1961) is computed between 110 wavelengths 465 and 619 nm for the CALITOO and 440 and 670 nm for the HHC, and between 440 and 675 nm for the AERONET.

Satellite data
The Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol products (Remer et al., 2005(Remer et al., , 2008 et al., 2016). The MODIS AOD is also used in operational data assimilation for weather forecast (Benedetti et al., 2009;Lynch et al., 2016). We have used the daily MODIS AOD at 550 nm from AQUA satellite (namely MYD08_D3) from 2003 to 2018. The spatial resolution of the MODIS product is 1 • × 1 • . We use the product named AOD_550_Dark_Target _Deep_Blue_Combined_Mean and Deep_Blue_Angstrom_Exponent_Land_Mean from the version 6 ( Levy et al., 2013) of the MODIS processing algorithm, which is a combination of the "Dark target" (Levy et al., 2010) and "Deep blue" (Sayer 120 et al., 2013) methods. For the purpose of satellite validation, the satellite AOD and AE of the nearest cell to the photometer location are extracted. We have adopted the evaluation metrics proposed by Sayer et al. (2014) including the linear correlation coefficient, the median bias, the root mean square error, the mean absolute percentage error, and the fraction of data falling within the MODIS expected error (EE) given by EE = ±0.05 + 0.15 × AOD. . Particles were collected on 47 mm diameter filters (quartz and PTFE filter types) at a flow rate of 5 L/min. Samplers were equipped with a PM2.5 mini Partisol impactor. PTFE filter were weighted before and after the sampling with a microbalance Sartorius MC21S. PM2.5 mass concentrations are estimated from the mass load on the filters and the total volume of air sampled measured by a GALLUS-type G4 gas meter. Exposure duration of the filter is one week. We have used the PM2.5 130 weekly observations collected at the urban site named "traffic" in Cotonou and the mean value at the two urban sites named "traffic" and "landfill" in Abidjan. In section 4, the AOD and AE are weekly averaged for the sake of comparison between PM2.5 and sun photometer observations.

Daily statistics
A total of 2323 handheld sun photometer observations (including data collected during the 2006 campaign) have been acquired.
Starting and ending dates are reported in  8 https://doi.org/10.5194/acp-2020-617 Preprint. Discussion started: 11 August 2020 c Author(s) 2020. CC BY 4.0 License. Table 2 gives the statistics of the regressions for each site and per instrument presented in Figure 4. We have then adopted a loglog representation on the scatter plots presented in Figure 4 as the AOD distribution has a significant right skewness (  The bias has a seasonal behavior being highest during the dry season between December and March. An underestimation of the MODIS AOD is then observed at maximum in January with an absolute bias of -0.33 (39% in relative) at the inland sites. Sayer et al. (2014) have already pointed out the possible differences in the "Dark Target" and "Deep blue" algorithms. It 185 appears from the Figure 6 in Sayer et al. (2014) that the dry to humid savanna transition zone in sWA is an area where large differences exist in both retrieval techniques during the dry season. Those differences can explain that the "Merge" product used in this study has a large bias during the dry season in the northern part for the inland sites. So the North-South AOD gradient in this area remains difficult to assess based on satellite products.

Comparison with MODIS aerosol products
The histogram of the Sun photometer and MODIS-derived AE are presented in Figure 5.  4 Aerosol type and relationship with surface concentrations AE depends on the aerosol size distribution and aerosol optical properties (Nakajima et al., 1996;Eck et al., 1999;Holben et al., 2001) and is commonly used to identify aerosol types (Léon et al., 1999;Kaskaoutis et al., 2009;Perrone et al., 2005). Aerosol with a lower value of AE than aerosol types having a size distribution dominated by the accumulation mode, like secondary 200 and combustion aerosols. The concurrent changes in AOD and AE help to distinguish generic aerosol types in sun photometer time series (Toledano et al., 2007;Verma, 2015). Mineral dust tends to increase atmospheric AOD and decrease AE (Hamonou et al., 1999) while biomass burning events tends to increase both AE and AOD (Eck et al., 2003). The thresholding in AOD and AE for aerosol type identification varies from one site to another and also depends on the distance from aerosol sources upwind the site (Verma, 2015;Benkhalifa et al., 2017).

205
In this paper we classify the daily observations according to the AE values using a simple statistical analysis. The whole sun photometer dataset is divided into 3 quantiles. The first third corresponds to AE≤0.45 and observations are labelled "coarse dust". The last third corresponds to AE≥0.80 and is labelled "urban-like". The data having 0.45<AE<0.80 falls into a "mixed" category. This rather crude classification enable to identify the main aerosol influence with a significant number of observations in each categories.  Table 3 presents the typology of the sites according to the aforementioned classification. The classification is given for the observations falling within the AOD IQR (see table 1). We also report the percentage of observations having AOD in the last quartile of the AOD distribution to highlight the contribution of the different aerosol types to the aerosol events. As expected the percentage of each catergory for a given site represents roughly one third of the data set. Two sites are less influenced by urban-like aerosols than the others, namely Save and Comoe. For all the sites excepted Koforidua, most of the days showing a 215 large AOD are associated with coarse dust events. Cotonou is also also more influenced bu Coarse Dust (39%) than Abidjan (28%). The northernmost sites are more affected by dust events and the influence of coarse dust on the AOD decreases from North to South and East to West.
As reported by Djossou et al. (2018), the changeover between the monsoon and the harmattan circulation leads to a drastic change in the aerosol type and stratification. The harmattan flow carries continental aerosols in the lowest part of the atmosphere 220 during the long dry winter season. During this period, the days with high AOD are often associated with an increase in the PM2.5 surface concentration. As a consequence, the correlation coefficient between AOD and PM2.5 is the highest during the dry long season (Djossou et al., 2018). Considering the whole dataset of PM2.5 and AOD (weekly basis) measured in Cotonou and Abidjan, the correlation between PM2.5 and AOD is significant with a correlation coeff. R=0.75, (N=105). The correlation can reach R=0.96 (N=6) during aerosol events observed from December 2015 to January 2016 in the heart of the dry season.

225
The variation in the PM2.5/AOD ratio is estimed as a function of the season and the aerosol type. The data set is divided The PM2.5/AOD ratio for coarse dust decreases from 49 ± 6 µg/m 3 /AOD to 29 ± 3 µg/m 3 /AOD between the long dry and long wet season. This change in the ratio reflects the seasonal change in the altitude of coarse dust transport. During the wet (long) season, the air masses are uplifted by the monsoon flow. PM2.5 concentrations remains moderate while AODs are 235 still significant due to aloft transport. Conversely the PM2.5/AOD ratio remains rather constant from the long dry to the short dry season, between 45 ± 4 µg/m 3 /AOD and 47 ± 16 µg/m 3 /AOD. This ratio increases in the short season and reaches 78 ± 12 µg/m 3 /AOD. This later period shows the largest uncertainties on the estimation of the PM2.5/AOD ratio due to moderate AODs leading to a less accurate regression.
13 https://doi.org/10.5194/acp-2020-617 Preprint. Discussion started: 11 August 2020 c Author(s) 2020. CC BY 4.0 License. q q q q q q q q q q q q q q q q q qq q q q qq q q q q q q q qq q q q q qq qq q q q q q q q q q q q q q qq q q q q q qq qq qq q q q q q q qq q q q qq q q q q qq q q q q q  in the urban-like PM2.5 corresponds to an average annual growth rate of 1.48% in agreement with the lower bound of the emission scenarii. However there is no evidence that the observed trend in our data is linked to an increase in the local city emissions. The phenomena can also be linked to the advection of biomass burning byproducts from central Africa and crossing the gulf of Guinea resulting from the zonal transport Flamant et al., 2018). In addition to satellite data, unraveling the causes and consequences of the changes in aerosol concentrations in this area of the world that is under a