Detection of a gas flaring signature in the AERONET optical properties of aerosols at a tropical station in West Africa

The West African region, with its peculiar climate and atmospheric dynamics, is a prominent source of aerosols. Reliable and long‐term in situ measurements of aerosol properties are not readily available across the region. In this study, Version 2 Level 1.5 Aerosol Robotic Network (AERONET) data were used to study the absorption and size distribution properties of aerosols from dominant sources identified by trajectory analysis. The trajectory analysis was used to define four sources of aerosols over a 10 year period. Sorting the AERONET aerosol retrievals by these putative sources, the hypothesis that there exists an optically distinct gas flaring signal was tested. Dominance of each source cluster varies with season: desert‐dust (DD) and biomass burning (BB) aerosols are dominant in months prior to the West African Monsoon (WAM); urban (UB) and gas flaring (GF) aerosol are dominant during the WAM months. BB aerosol, with single scattering albedo (SSA) at 675 nm value of 0.86 ± 0.03 and GF aerosol with SSA (675 nm) value of 0.9 ± 0.07, is the most absorbing of the aerosol categories. The range of Absorption Angstrӧm Exponent (AAE) for DD, BB, UB and GF classes are 1.99 ± 0.35, 1.45 ± 0.26, 1.21 ± 0.38 and 0.98 ± 0.25, respectively, indicating different aerosol composition for each source. The AAE (440–870 nm) and Angstrӧm Exponent (AE) (440–870 nm) relationships further show the spread and overlap of the variation of these optical and microphysical properties, presumably due in part to similarity in the sources of aerosols and in part, due to mixing of air parcels from different sources en route to the measurement site.


Introduction
Some of the reservoirs from which crude oil is explored from the Earth's crust contains natural gas. This natural gas is separated at flow stations in the several oil and gas fields in every oil producing region of the world. This gas could be put to meaningful use (for example, power generation, domestic, and industrial uses). Where the infrastructure, technology, and/or market to put the gas to meaningful use are not available, the gas is either flared or vented into the atmosphere. Considering the climate forcing potential of methane, a major component of the gas, flaring is the better option. See Fawole et al. [2016] for a better understanding of the gas flaring process and its environmental impact. Gas flaring is the process of disposing of natural gas by combustion in an open flame in the open atmosphere, using a burner tip designed specifically for that purpose, in the course of routine oil and gas production operations. Flaring is most often associated with Nigeria and Russia, but it still takes place even in more developed economies, such as the case of the North Dakota Bakken shale region. The intensity of gas flaring and specifics of atmospheric transport can combine to enhance the role of gas flaring emissions over very large areas.
Gas flaring remains a significant source of air pollutants such as aerosol sulfate, black carbon (BC, colloquially "soot"), carbon monoxide (CO), polyaromatic hydrocarbon (PAH), and volatile organic compounds (VOCs), especially in those regions without the technical infrastructure needed to trap and process natural gas associated with oil reserves [Fawole et al., 2016]. Tracking these aerosols from the source requires an understanding of the chemical character of the source and the prevailing meteorological conditions [Eck et al., 2010;Giles et al., 2012;Rozwadowska et al., 2010]. With atmospheric residence times of 5-8 days, gas flaring associated aerosol, especially black carbon (BC), is a good tracer for gas flaring emissions [Shindell et al., 2008]. This study aims to assess and quantify the contributions of intense gas flaring activities in the Nigeria oil field FAWOLE ET AL.
OPTICAL PROPERTIES OF AEROSOLS 14,513 to regional aerosol loadings. Gas flaring is a prominent contributor to atmospheric aerosol loading, on both local and regional scales, in leading oil producing nations of the world, including Russia, Nigeria, the USA, Iran, and Iraq [Doherty et al., 2014;Fawole et al., 2016].
Aerosol in the atmosphere can be classified in many different ways; here it is useful to consider fineand coarse-absorbing classes, based on their radiation absorption potential and Mie-scattering particle-size equivalence [Mielonen et al., 2009]. The spectral dependence of aerosol optical and microphysical properties derived from remote sensing measurements of aerosol properties have been used to establish dominant aerosol signals [e.g., Giles et al., 2012;Russell et al., 2010]. Aerosol nature and properties vary with region; so adequate knowledge of major sources of aerosol in a region should enhance identification of dominant aerosol types at a measurement site in the region. Back trajectory analysis and adequate knowledge of the sources contributing to aerosol loadings in a region can, therefore, give a clearer understanding of the properties of the aerosol from the dominant sources [Milinevsky et al., 2014].
The Aerosol Robotic Network (AERONET) consists of more than 400 sites distributed around the globe. This network of automated Sun photometers has provided retrievals of aerosol properties for more than 20 years at some sites [Holben et al., 1998;Holben et al., 2001]. Many studies have utilized aerosol retrievals from the AERONET data set to analyze optical properties and size distribution and, hence, composition of aerosols in different regions of the world [e.g., Giles et al., 2012;Kim et al., 2011;Milinevsky et al., 2014;Russell et al., 2010;Sato et al., 2003].
Previous studies have classified the Ilorin AERONET site in West Africa, described in more detail below, as either a predominant dust [Lee et al., 2010;Pandithurai et al., 2001;Pinker et al., 2010;Smirnov et al., 2002] or mixed aerosol [Eck et al., 2010;Giles et al., 2012] site. The location of Ilorin ( Figure 1) as the boundary city between the densely populated and industrialized Monsoonal forest region of the south and Savanah region of the north of Nigeria places it at a strategic location for aerosol-climate interactions. The prevailing winds are the dry northeasterly "Harmattan" (NEH) and moist southwesterly monsoon (SWM) winds. The SWM wind is the predominant wind during the WAM season. NEH brings dry Harmattan dust from the Sahara and Sahel region, usually between late November and early February, while the SWM brings in moist cool air that gives rain [Sultan and Janicot, 2003] but which may also carry urban and industrial aerosols from the south.
Gas flaring, the disposal of natural gas through stacks in an open-air flame, has been identified as a substantial, distinctive, and persistent source of atmospheric aerosol and pollutant trace gases [Fawole et al., 2016;Huang et al., 2015;U.S. Environmental Protection Agency, 2012]. Atmospheric pollutants emitted from gas flaring include CO, SO 2 , PAH, VOCs, and black carbon (BC). To the south of the Ilorin AERONET site is the gas flaring region in Nigeria with over 300 active flare sites [Anejionu et al., 2015a;Elvidge et al., 2015] and an estimated annual average volume flared of 15 bcm (billion cubic meters). In the region, gas flaring is a daily routine at many of the sites, owing to the large amounts of natural gas produced, and inadequate piping to facilitate bringing the product to the market. Nigeria, presently ranked the second leading gas flaring nation of the world, is estimated to flare about 25% of its annual gas production [Ite and Ibok, 2013]. Figure 1 shows the spatial distribution of active flare sites in the Niger Delta, Nigeria, and location of the Ilorin AERONET site. In 2012, of the 325 active flare sites identified in the Nigeria oil field, 97 (~30%) ranked among the top 1000 largest flares identified globally. Anejionu et al. [2015b], in their analysis of the multiyear spatial dispersion of gas flaring gaseous emissions from the Niger Delta region, classified the airshed of more than 70% of the Niger Delta as moderate to high-pollution risk areas. Giwa et al. [2014], using emission factors from literature, estimated that gas flaring contributes about 10 Gg of black carbon (BC) to ambient BC concentration in the Niger Delta annually.
In the present study, using a combination of trajectory analysis and knowledge of dominant aerosol sources in the region, we The AERONET site is located in Ilorin, Nigeria (8.32°N, 4.34°E) ( Figure 1). Ilorin has a tropical wet and dry/savannah climate. Based on the Köppen-Geiger climatic classification, Ilorin can be classified as equatorial savannah with dry winter (Aw). In Ilorin, the annual mean rainfall and temperature are about 1220 mm and 26.5°C, respectively [Olaniran, 1982;Tunde et al., 2013]. The region has two big cities-Lagos and Oyo, city regions with populations of 17 and 5 million, respectively-and the gas flaring region (Niger Delta) to the south, the Sahel and Sahara region to the north and major sources of biomass burning emissions in West Africa to the east and west. The movement of the Intertropical Convergence Zone (ITCZ) and Intertropical Front (ITF) determines the preonset and onset of the West African Monsoon (WAM) in the region. Biomass burning emissions are intense in the region between November and March [Roberts et al., 2009], a period during which there is also significant dust transport from the Sahel and Sudan region. Saharan dust is predominant between late November and January, a period of intense Harmattan conditions. Depending on the season and prevailing meteorological conditions, there is an influx of urban and industrial aerosols from the south. We expect this influx of urban-industrial air to predominate during the WAM months, between April and October. We hypothesize that aerosols from each of these sources exhibit statistically significant differences in their optical-and, hence, microphysical-properties.   and pressure) are assigned as attributes to the particle in the trajectory. Detailed technical descriptions of the UGAMP trajectory model can be found in Methven [1997] and Methven et al. [2001].
Nine (9) backward trajectories spaced 0.05°× 0.05°(i.e., 5 × 5 km) apart on a 0.1°× 0.1°grid were released 6 hourly, that is, at 0000, 0600, 1200, and 1800, at 900 hPa (i.e., approximately 1.5 km altitude) on each of the days for which the calculations were done. It must be noted that these trajectories, in general, change altitude as a function of transit time: 900 hPa is the arrival pressure at the AERONET site. The choice of 7 day back trajectory length is due to the atmospheric lifetime of between 5 and 9 days estimated for black carbon (BC) and particulate organic matter [Cooke et al., 1997;Cooke and Wilson, 1996;Koch et al., 2009;Stier et al., 2006], both of which are major constituent of aerosol in the study area. Desert dust is also a prominent aerosol constituent, especially during the non-WAM months.
The clustering of the aerosol signals by their trajectory paths enables a consideration of the major sources of aerosol in the region as shown in Figure 2. See earlier studies [e.g., Omar et al., 2005;Russell et al., 2014] for more on the clustering of aerosol parameter using AERONET data. In this study, the classification was based solely on the trajectory coordinates, considering whether the trajectory had passed through the gridded regions of one or more of the major aerosol sources where they are suggested to have picked up aerosol signatures peculiar to such sources. The classification algorithm developed and used in this study allows the large data set (10 years of trajectories) to be processed automatically avoiding manual inspection of trajectory maps.
When at least six of the nine trajectories released at a point in time pass through one of the source region, that region is considered to be the dominant source of the aerosol signal measured. When there are an almost equal number of trajectories from two sources in the four sets of 6-hourly trajectories released on a day, the aerosol signal recorded on that day is regarded as a mixture of the two most dominating sources. In this way, eight clusters were created: BB, DD, UB, GF, DD-BB, DD-UB, GF-UB, and GF-DD. BB-UB and BB-GF clusters occurred less than 1% of the time and are therefore not investigated further. In this study, we analyze the optical properties of BB-, UB-, GF-dominant and "pure" DD clusters as well as their climatologies. Box plots presenting variation of optical and microphysical properties of the "mixed-source clusters" are available in the supporting information (see Figure S1). Because AERONET measurements observe the aerosol through the depth of the atmosphere, differential advection with height can bring air of different origins above the site simultaneously. We do not attempt to quantify this and limit our study to finding dominant sources in air arriving close to the surface (900 hPa).

AERONET Data
AERONET network provides Sun photometer measurements of AOD at up to eight wavelengths between 340 and 1020 nm. The network has been in operation since mid-1990s [Holben et al., 1998] and currently with over  [Chubarova et al., 2016]. It also measures the angular distribution of sky radiance at four wavelengths (0.44, 0.67, 0.87, and 1.02 μm). An inversion algorithm is used for the retrieval of optical and microphysical properties of atmospheric aerosols, such as volume size distribution, asymmetric parameter (g), complex refractive index (m), and single scattering albedo (SSA) at 0.44, 0.67, 0.87, and 1.02 μm [Dubovik et al., 2002;Eck et al., 2010]. For sites like Ilorin where differences in the diurnal variation of aerosol are very pronounced, the use of monthly averages of aerosol parameters will provide only highly generalized estimates of the optical and microphysical properties of aerosol at such a site.
The Angstrӧm Exponent (AE), a measure of the wavelength (λ) dependence of aerosol optical depth (τ), indicates the dominant aerosol size because the spectral shape of the extinction is inversely related to the particle size [Eck et al., 1999;Schuster et al., 2006]. The wavelength dependence of scattering as quantified by AE (α) is given by equation (1): Level 1.5 of Version 2 AERONET retrievals were used in the analyses of the optical and microphysical properties of the clusters because Level 2 data, especially inversion products, were not sufficiently available. Level 2 data of aerosol parameters from the Ilorin AERONET site were used to analyze the dynamics of the seasonal variation of fine-mode fraction (FMF), AOD, and AE (470-870 nm) [O'Neill et al., 2003]. Although, Level 1.5 data are cloud screened, to further enhance the quality of the Level 1.5 data used in this analysis, we have included only Level 1.5 data for which Level 2 size distribution data are available. This restriction ensures that only Level 1.5 data for which solar zenith angle > 50°are used in the analysis [see Bond et al., 2013;Dubovik et al., 2002]. According to , SSA values get worse when AOD (440 nm) ≤ 0.2. Hence, we have used a threshold value of AOD (440 nm) > 0.2 in this analysis. The quality control approach adopted, as stated above, implies that we have applied the same quality control check for Level 2 retrievals except that we have relaxed the AOD (440 nm) requirement from 0.4 to 0.2.
The variation of absorption aerosol optical thickness (τ abs ) with wavelength (λ) can be approximated with a logarithmic wavelength dependence to infer the Absorption Angstrӧm Exponent, AAE, (α abs ), as given in equation (2). Similarly, as given in equation (3), the Extinction Angstrӧm Exponent, EAE, (α ext ) is obtained from the spectral dependence of extinction aerosol optical thickness (τ ext ) with wavelength (λ) on a logarithmic scale.
The spectral single scattering albedo (SSA) of an aerosol, as given in equation (4), is the ratio of scattering to extinction (the sum of absorption and scattering). SSA depends on the strength of the absorption of the aerosol sources and the aging of the aerosol during atmospheric transport. For atmospheric aerosols, the value of SSA and AOD primarily determines the sign and magnitude of its radiative forcing, respectively.

Climatic Condition at the Study Site
The predominant seasons at the site are governed by the movement of the Intertropical Convergence Zone (ITCZ) and Intertropical Front (ITF). The ITF is the northern limit of the southwesterly winds of the monsoon. The preonset stage of the West African Monsoon (WAM), usually between May and June, is characterized by the movement of the ITF further north to 15°N which brings in the first period of rain. A further shift of the ITF to its northernmost position around 20°N takes place between July and August, a period which signifies the onset of WAM [Sultan and Janicot, 2000;Sultan and Janicot, 2003]. At the onset of the WAM, there is an abrupt shift in the latitudinal position of the ITCZ from a location of 5°N in May-June to a location at 10°N in July-August.  The monthly average AOD pattern is shown in Figure 3a for 440, 675, and 870 nm. For the analysis of the daily variation of AOD and AE, this 5 year period (2005)(2006)(2007)(2008)(2009)) was chosen because there were very few missing data during these years. At the Ilorin AERONET site, there are frequent breaks in data availability probably due to adverse meteorological conditions, for example, persistent cloud cover or instrument breakdown.

10.1002/2016JD025584
During the WAM months, the several spikes observed in the daily AOD data are most likely due to local dust rather than dust transport from the Sahara. In contrast to cities in the developed economies, there is a dominance of the coarse-mode fraction in the atmospheric aerosol loading in major Nigerian cities. This has been attributed to frequent and persistent resuspension of dust from the large expanse of unpaved roads [Akinlade et al., 2015;Obioh et al., 2013;Owoade et al., 2013].

Seasonal Dynamics of Aerosol Absorption: Single Scattering Albedo (SSA)
Single scattering albedo (SSA) of atmospheric aerosols is an important parameter in estimating the radiative effect of the aerosol. Depending on the surface reflectance (albedo), SSA determines the sign of the radiative effect, hence its cooling/warming potential; it ranges from 0 (zero) for a purely absorbing aerosol to 1 for a purely scattering aerosol.
During the WAM months, the average monthly SSA decreases with wavelength (see the green ellipses in Figure 5), and slightly increases with wavelength during the non-WAM months (DJF). This decrease, between 5.2 and 5.7%, is attributable to the fact that scattering often decreases faster than absorption with wavelength for carbonaceous particles [Bergstrom et al., 2007]. SSA is also highest during these months for the 5 year period shown in Figure 5, due presumably to increased concentrations of less absorbing particles in the aerosol influx from the urban and gas flaring (industrial) regions downwind of Ilorin. SO 2 , a precursor gas for atmospheric sulfate, is a major pollutant from gas flaring of "sour" natural gas sources [Fawole et al., 2016]. Mixtures of sulfate and BC (carbonaceous particles) have been associated with high SSA values [Lim et al., 2014].  aerosol loadings, measured remotely by Sun photometers, in terms of their dominant compositions, nature, and potential origin [Dubovik et al., 2002;Milinevsky et al., 2014;Russell et al., 2010].

Optical and Microphysical Properties of Aerosol Clusters
Of the total number (N = 801) of aerosol signatures assigned to the various clusters, BB-, UB-, GF-dominant and pure DD account for 184 (23%), 212 (26.5%), 167 (20.8%), and 238 (29.7%), respectively. This suggests that gas flaring (GF) is a major contributor to the aerosol signature in the region, especially during the WAM months. The number of trajectories classified into each of the "mixed clusters" is as follows: UB-GF (N = 145), GF-DD (N = 55), DD-UB (N = 63), and BB-DD (N = 119). The mixed clusters are not discussed further in this study.

Desert Dust (DD) Aerosol Cluster
Desert dusts are predominant in the region with the coming of the NEH wind usually between late November and February when both the ITCZ and ITF are nearer the equator. The DD cluster is associated with high AOD (median = 1.02; Q1 and Q3 are 0.72 and 1.32, respectively): the association of high AOD and low AE is typical of desert dust [Toledano et al., 2007]. Table 1a, for the DD cluster, the range of values for the AAE, and AE are 1.64 ≤ α abs ≤ 2.34 and 0.19 ≤ α ext ≤ 0.49, respectively. These values corroborate findings from similar sites [Giles et al., 2012;Russell et al., 2010]. The absorption potential of desert dust as shown in the SSA 440 values ω = 0.90 AE 0.03 is determined by the hematite (iron oxides) content of desert dust, which is a strong determinant of the radiative  effect of the aerosol. SSA 440 is given here because iron oxide absorbs much more in the shorted wavelength. As expected for aerosol in this cluster, the AE values show a dominance of the coarse-mode fraction. Hence, this cluster can be referred to as coarse absorbing.

Biomass Burning (BB) Aerosol Cluster
Aerosol in this cluster is predominant in the months before the preonset of the WAM, i.e., between November and March. In West Africa, March is usually the peak of the dry season and, as such, intense agricultural burning takes place between November and March to prepare the land for planting during the forthcoming rainy season. Intrusion of desert dust is also a regular occurrence within this period. The trajectory clustering enabled us to identify days of dominance of both aerosol signatures with these months (November-March). There are days of an almost equal contribution of trajectories from these sources (DD and BB) to the back trajectory analysis. Such days are categorized as a mixture from both sources, that is, DD-BB.
The BB cluster is characterized by high AOD and high AE values; the high AE value is evident in the EAE values shown in Table 1a. The high carbon content of biomass burning aerosols is responsible for its high absorbing nature as expressed by its SSA 675 values (ω = 0.86 AE 0.04), the lowest of the four clusters. The ranges of the AAE and AE values are 1.19 ≤ α abs ≤ 1.71 and 0.73 ≤ α ext ≤ 1.23, respectively. This cluster, similar in value to other biomass burning sites in the literature, has a relatively high FMF value, while the AE value for this cluster is slightly lower than expected. This probably relates to the nature of vegetation burned, the aging of aerosol during transport to the measurement site, or a slight contribution from dust, mixed-in or aloft.

Urban Aerosol (UB) Aerosol Cluster
Urban (UB) aerosol signatures are predominant in the trajectory classification between May and October; it is often intense between June and September (JJAS). This is due to northward shift of the ITCZ and ITF (as discussed in section 2.4) which allows inland influx of moist SWM wind that brings rain. This wind brings pollution laden aerosol from the large cities, Lagos and Oyo, with high human population densities.
The UB cluster has an average SSA 675 value of 0.92 (AE0.04), that is, it is relatively more absorbing than the DD cluster due to the presence of carbonaceous particles from combustion processes and vehicular emissions. The range of AAE values, 0.83 ≤ α abs ≤ 1.59, is very similar to observations from similar urban sites [Giles et al., 2012]. The average AE value, compared to similar sites in literature, is lower by a factor of 2.5 which suggests the dominance of coarse-mode particles. This lower AE values are attributable to dust resuspension from the predominantly unpaved and damaged paved road networks that lie across major cities in Nigeria. Several in situ PM measurements across various site classifications (residential, marine, heavy traffic, and industrial) in major Nigeria cities have reported the dominance of the coarse-mode (PM 2.5-10 ) PM fraction [Akinlade et al., 2015;Ezeh et al., 2012;Obioh et al., 2013;Owoade et al., 2013].

Gas Flaring (GF) Aerosol Cluster
The gas flaring region, about 500 km south of the AERONET site, contains no fewer than 300 flaring sites scattered around the Niger Delta, a land mass of about 70, 000 km 2 [Anejionu et al., 2014;Elvidge et al., 2015]. Onyeuwaoma et al. [2015], studying aerosol loading patterns in major Nigerian cities from Moderate Resolution Imaging Spectroradiometer and OMI data, identified gas flaring from the oil and gas fields in the Niger Delta area as a contributor to elevated concentrations of carbon monoxide (CO) observed in cities near the flaring region. Wind trajectories from the gas flaring region are captured in the trajectory analysis predominantly during the WAM months.
Of the four dominant clusters identified, aerosols in the GF cluster have the lowest average AAE value of 0.98. This value, which is very close to the theoretical AAE value of 1 associated with BC [Bergstrom, 1973;Bohren and Huffman, 1983], is attributed to the BC component of gas flaring aerosols. There is some overlap in the AAE values of the BB and GF clusters, probably due to both having substantial carbonaceous particle content. Although, as shown in Figure 7, the AE values for the BB and GF clusters overlap; the statistical comparison of the sample distributions shows that the means are significantly different (see supporting information). For the GF cluster, the ranges of the AAE and AE values are 0.72 ≤ α abs ≤ 1.24 and 0.77 ≤ α ext ≤ 1.37, respectively. See the supporting information ( Figure S2) for box plots showing the spread and overlap of the different properties for the four clusters. The average SSA 675 (ω = 0.9 AE 0.06) value for the GF cluster shows that it is more absorbing than the UB-dominant aerosol cluster. Like BB-dominated aerosol cluster, aerosols in this cluster can be referred to as fine absorbing. Figure 7 shows the spread of the AAE and AE of the four aerosol clusters identified. DD-dominated aerosols are unique in terms of AAE and AE, while there are varying degrees of overlaps in those of UB, BB, and GF dominated aerosols. The statistical comparison of the mean values of AAE and AE for the four aerosol clusters is presented in the supporting information.
To examine the distribution of the aerosol parameters, Table 1b presents the median values of the optical and microphysical properties of the four aerosol clusters identified. The skewness of the distribution of these properties in all the clusters is very mild as shown by the closeness of the mean and median values of the aerosol parameters as shown in Tables 1a and 1b.

Conclusions
Earlier studies have classified the Ilorin AERONET site as dominated by dust or a mixture of dust and biomass burning aerosols. Analysis presented here for 10 years of aerosol data from this site allows, for the first time, the identification and analysis of contribution of dominant sources to the mixed aerosol loading at the site. The aerosol optical depth (AOD) and Angstrӧm Exponent (AE) vary significantly during the year due to a significant change in the nature and composition of dominant aerosols at the site. The months before the West African Monsoon (WAM) are characterized by aerosol with high AOD and low AE, while the WAM months witness aerosols with low AOD and high AE.
Back trajectories classification for the study period (2005-2009 and 2011-2015) and known major source contributors to aerosol loading in the region were used to create three aerosol-dominant (BB, UB, and GF) and one pure DD source-related clusters from AERONET retrievals, which vary significantly in optical and microphysical properties. Biomass burning and desert dust-dominant aerosols are prominent in the pre-WAM months, while gas flaring and urban aerosols dominate the WAM months.