Black carbon aerosols are the second largest contributor
to global warming while also being linked to respiratory and cardiovascular
disease. These particles are generally found in smoke plumes originating
from biomass burning and fossil fuel combustion. They are also heavily
concentrated in smoke plumes originating from oil fires, exhibiting the
largest ratio of black carbon to organic carbon. In this study, we
identified and analysed oil smoke plumes derived from 30 major industrial
events within a 12-year timeframe. To our knowledge, this is the first study
of its kind that utilized a synergetic approach based on satellite remote
sensing techniques. Satellite data offer access to these events, which, as
seen in this study, are mainly located in war-prone or hazardous areas. This
study focuses on the use of MODIS (Moderate Resolution Imaging
Spectroradiometer) and CALIPSO (Cloud-Aerosol Lidar and Infrared Pathfinder
Satellite Observations) products regarding these types of aerosol while also
highlighting their intrinsic limitations. By using data from both MODIS
instruments on board Terra and Aqua satellites, we addressed the temporal
evolution of the smoke plume while assessing lidar-specific properties and
plume elevation using CALIPSO data. The analysis method in this study was
developed to better differentiate between oil smoke aerosols and the local
atmospheric scene. We present several aerosol properties in the form of
plume-specific averaged values. We believe that MODIS values are a
conservative estimation of plume aerosol optical depth (AOD) since MODIS algorithms rely on general
aerosol models and various atmospheric conditions within the look-up tables,
which do not reflect the highly absorbing nature of these smoke plumes.
Based on this study we conclude that the MODIS land algorithms are not yet
suited for retrieving aerosol properties for these types of smoke plumes due
to the strong absorbing properties of these aerosols. CALIPSO retrievals
rely heavily on the type of lidar solutions showing discrepancy between
constrained and unconstrained retrievals. Smoke plumes identified within a
larger aerosol layer were treated as unconstrained retrievals and resulted
in conservative AOD estimates. Conversely, smoke plumes surrounded by clear
air were identified as opaque aerosol layers and resulted in higher lidar
ratios and AOD values. Measured lidar ratios and particulate depolarization
ratios showed values similar to the upper ranges of biomass burning smoke.
Results agree with studies that utilized ground-based retrievals, in
particular for Ångström exponent (AE) and effective radius
(
Atmospheric aerosols are chemically complex mixtures of solid and liquid particles dynamically suspended in air. They originate from both natural and anthropogenic emissions. More common naturally occurring aerosols can be observed in the form of fog, dust, sea salt spray, biological exudates and grey smoke (biomass burning). Haze, smog and black smoke are typically a result of industrial and transportation activities (Stocker et al., 2014; Wei et al., 2020). Distinct species such as black carbon (BC), organic carbon (OC), sulfates, nitrates, trace elements, sea salt, mineral dust and biological matter suffer atmospheric alteration, resulting in different combinations of compounds. Defining aerosol types is a difficult task as they possess a large degree of variance in composition and concentration due to different atmospheric residence times, dry deposition and wet scavenging, emission rates and sources, transport trajectories, and seasonal variability (Dutkiewicz et al., 2009; Li et al., 2015; Samset et al., 2018).
Health effects associated with both short- and long-term exposure to aerosol
have been widely documented in scientific literature (Brauer
et al., 2015; Laumbach and Kipen, 2012; Zhang and Batterman, 2013; Pascal et
al., 2013; Lee et al., 2014; Guarnieri and Balmes, 2014; Zhang et al.,
2017). Aerosols have been linked to respiratory and cardiovascular diseases
due to fine particulate matter (PM
Global circulation of aerosols is a known transport vector of minerals and nutrients to the biosphere (McTainsh and Strong, 2007; Maher et al., 2010; Lequy et al., 2012). Aerosols have a direct effect on radiation distribution by scattering, absorbing and emitting light through the atmosphere. In addition, they can affect the climate system through indirect effects acting as cloud condensation nuclei, impacting cloud lifetime and properties, atmospheric stability and precipitation factors (Popp et al., 2016; Samset et al., 2018; Stocker et al., 2014). They can disrupt circulation patterns, impact air temperatures and severe weather systems (Fan et al., 2016), reduce visibility (Deng et al., 2012; Wang et al., 2015), and lead to ozone depletion (Popp et al., 2016).
Because of their complex influence on the environment and climate system, assessing aerosol key parameters is essential for any atmospheric study. Aerosol optical depth (AOD), the extinction vertically integrated throughout the atmospheric column, is strongly correlated to PM concentrations. Together with other properties such as Ångström exponent (AE), single-scattering albedo (SSA), size distribution and vertical distribution, we can better describe their atmospheric impacts. Between ground stations and spaceborne observations, satellite remote sensing offers a more comprehensive global view of aerosols. Since the 1970s onwards, there has been a significant number of satellite sensors used successfully for retrieving AOD and other aerosol properties (Li et al., 2015, 2016; Dubovik et al., 2019; Schutgens et al., 2020; Wei et al., 2020; Sayer et al., 2020). When choosing between different aerosol products, one must take into account the wide variety of sensors and their characteristics such as spatial, temporal and spectral resolutions; single or multi-view retrieval methods; intensity or polarimetric design; and different retrieval algorithms (Wei et al., 2020; Fan and Qu, 2019; Sogacheva et al., 2020; Li et al., 2020). In addition to sensor characteristics, other factors such as cloud coverage, surface type, aerosol models and retrieval algorithms contribute to overall retrieval uncertainties (Wei et al., 2020; Li et al., 2015; Virtanen et al., 2018). Most sensors can retrieve a wide variety of aerosol properties; however they rely on inversion techniques and complex radiative transfer computations (Schutgens et al., 2020).
Smoke aerosols are primarily composed of two distinct carbonaceous species:
BC, highly absorbent in all visible wavelengths, and OC, highly scattering of
solar radiation (Ramanathan and Carmichael,
2008; Dutkiewicz et al., 2009). Recent studies suggest black carbon (BC) is
the second largest contributor to global warming after CO
The most abundant source of atmospheric data on oil smoke plumes was gathered from the Kuwait oil fires in 1991. An estimated 700 oil wells were set on fire while smoke plumes engulfed large areas of the Gulf of Kuwait region (Cahalan, 1992). The amount of burned oil was estimated between 1.2 and 7.5 million barrels per day (Sadiq and McCain, 1993). Satellite images of visible smoke were first acquired on 9 February spanning until November when the last fires were extinguished (Draxler et al., 1994; Limaye et al., 1992). Several international research teams conducted extensive field campaigns, concentrating their efforts on atmospheric, environmental and health-related issues focusing on the potential impacts on global climate (Sadiq and McCain, 1993; Husain, 1995; World Meteorological Organization, 1993). Studies on health effects related to the Kuwait oil fires suggest that smoke exposure led to acute respiratory illnesses with some suspecting long-term effects (Etzel and Ashley, 1994; Brain et al., 1998; Smith, 2002; Lange et al., 2002; Kelsall, 2004; Heller, 2011; Barth et al., 2016). Valuable atmospheric data were also collected from smaller events such as oil depot fires, most notably the Buncefield incident on 11 December 2005. A number of explosions led to a large fire engulfing 20 storage tanks until 15 December. The fire burned 58 000 t of fuel while injecting a large smoke plume above the boundary layer at 3000 m (Vautard et al., 2007; Health and Safety Executive and Buncefield Major Incident Investigation Board (Great Britain), 2008). An initial report on air quality concluded that the smoke plume remained aloft over cold and stable atmospheric layers, thus reducing the potential impacts at ground level (Targa et al., 2006). Health studies related to the event concluded no long-term impacts on people exposed to the smoke; however acute respiratory symptoms were reported (Hoek et al., 2007; Morgan et al., 2008).
Oil fire and biomass burning (BB) smoke plumes significantly differ in
One objective of this study is to highlight the importance of satellite remote sensing techniques in identifying these types of events. As opposed to ground-based data, satellite data offer access to remote areas all over the globe, which would otherwise be very difficult to achieve. As seen in Table 1, oil installations may be situated in desert areas, at sea or in secluded locations far away from air quality monitoring stations or AERONET (AErosol RObotic NETwork) sites (Holben et al., 1998). In addition to this advantage, a synergistic approach using different types of satellite instruments can offer three-dimensional space coverage. While in situ ground stations and modelling tools are viable options for smoke plume research, these methods have limitations in areas prone to armed conflicts or posing high health risks. Out of the aforementioned events, only event 10 was analysed by different techniques, as seen in local AERONET data (Sect. 3.4). It goes without saying that retrieving optical and microphysical properties of petrochemical burnings may be challenging in most cases even with this approach. This study will focus on the use of MODIS and CALIPSO aerosol products regarding these types of aerosol while also highlighting their limitations. By using data from both MODIS instruments on board Terra and Aqua satellites, we addressed the temporal evolution of the smoke plume while assessing lidar-specific properties and plume elevation using CALIPSO data. The low number of studies on petrochemical smoke plumes, especially in the last decade, further encourages us to address these issues. While biomass burning and industrial haze are abundantly discussed in scientific literature, the same cannot be said for petrochemical smoke plumes resulting from major technological accidents. To our knowledge, we have not identified any similar studies focused specifically on retrieving aerosol properties from major petrochemical accidents by using synergistic satellite techniques.
Major industrial events leading to observable smoke plumes seen in MODIS RGB images.
This section summarizes a collection of events ranging from 2008 to 2019 that were successfully identified by satellite remote sensing techniques. Table 1 also provides the coordinates and the number of MODIS observation for each of the events covered in this study.
Events similar in nature to the Kuwait oil fires took place in northern Iraq as oil fields at Qayyarah and Najma were intentionally set ablaze by Islamic State of Iraq and Syria (ISIS) militants in an attempt to deter coalition air strikes. The first fires were detected east of Baiji in early January 2016. Other oil wells burned intermittently from May to June close to Mosul and Kirkuk. The bulk of smoke plumes were observed largely between June and November; however smoke plumes were continuously detected from the Qayyarah oil fields for a total of 225 d ranging from 13 June 2016 to 27 March 2017. As a result of these events, an estimated 1.33 million barrels of oil were burned (Bulmer, 2018; Tichý and Eichler, 2018). Residents south of the Qayyarah oil fields were exposed for 103 d to smoke plumes. Short-term health effects were reported, especially for patients with pre-existing respiratory conditions (Bulmer, 2018).
The Gulf of Sidra has seen extensive episodes of smoke plumes as oil terminals at As Sidr and Ra's Lanuf, Libya, have been repeatedly set on fire over the course of a decade. These events have been captured by MODIS sensors through the last decade all the way since 2008. All events were characterized by dark plumes, suggesting high contents of BC. On 19 August 2008 a tank fire erupted in Fiba tank farm at Ra's Lanuf after workers failed a maintenance operation (Piafom, 2018; The Telegraph, 2011). The fire lasted 9 d, during which smoke plumes could be detected from 19 to 22 August. In March 2011 the terminal was struck by air artillery in the battle of Ra's Lanuf as the country was engaged in civil war (BBC, 2011; The Christian Science Monitor, 2011; The Guardian, 2011). Smoke plumes were visible on 12 and 14 March. In December 2014 the tank farm at As Sidr oil terminal was struck by rockets as rebels fought to seize control of the city port. Seven storage tanks where engulfed in flames, burning 1.8 million barrels of crude oil (Reuters, 2014; BBC, 2014; Al Jazeera, 2014). Smoke plumes covered large areas of the Gulf of Sidra between 26 and 30 December and could be seen as far east as Timimi and Crete. In January 2016 both tank farms at As Sidr and Ra's Lanuf were struck by Islamic State militants. On 5 January As Sidr suffered five tank fires, while two tanks were hit at Ra's Lanuf, amounting to 850 000 barrels of oil. A week later IS militants struck oil infrastructure connecting Ra's Lanuf terminal to other installations in the area (Tichý and Eichler, 2018; Tichý, 2019). A second attack at Ra's Lanuf tank farm was conducted on 21 January (Business Insider, 2016). Throughout the month, extremely dense smoke plumes could be seen in the region from 5 January all the way to 23 January. The most recent incident at Ra's Lanuf took place in June 2018 when rival armed groups clashed. The fire which started on 14 June was contained at two storage tanks before it was extinguished several days later (Reuters, 2018). Visible smoke plumes were detected on 17 and 18 June (Bellingcat, 2018).
The MODerate resolution Imaging Spectroradiometer (MODIS) is a passive
remote sensing instrument on board NASA's Earth Observing Satellites (EOS).
The instrument has been collecting climate-related data, including aerosol
products, since 2000 from Terra and 2002 from Aqua satellite platforms. To
achieve a vast catalogue of products, MODIS uses its wide spectral range, 36
channels between 0.41 and 14.5
Herein we will summarize the aerosol retrieval algorithms Dark Target (DT) over land and ocean (Kaufman et al., 1997; Tanré et al., 1997; Levy et al., 2007a, b, c, 2013; Remer et al., 2005, 2008, 2013) and Deep Blue (DB) over land (Hsu et al., 2004, 2006, 2013), with some emphasis on the atmospheric parameters used in the construction of look-up tables (LUTs) and aerosol model selection as these properties/assumptions are crucial for proper AOD retrieval in oil smoke events.
The DT land algorithm is used over dark vegetated surfaces with low surface
reflectance. DT makes use of the “VIS to 2.1” relationship to distinguish
surface contributions to the top-of-atmosphere (TOA) reflectance, as
aerosols have a low absorbing and scattering effect in the shortwave infrared
(2.12
The DT ocean algorithm works in much the same way as the land algorithm,
although it requires masking sediments and filtering out strong glint areas.
It uses the spectral dependencies of six bands, 0.55, 0.65, 0.87, 1.24, 1.63
and 2.12
Aerosol properties within the LUT such as size distribution parameters, refractive indexes and SSA are crucial for proper aerosol typing and subsequent AOD retrieval. LUT information and model selection are critical for deriving other optical properties, such as Ångström exponent (AE), which can also be used to describe size distribution.
The DB algorithm was developed to retrieve AOD over arid, semi-arid and
urban areas where surface reflectance values are higher than those over Dark
Target regions. The principle behind the algorithm suggests that surface
reflectance in these areas shows higher values in red and near-infrared bands
and lower values in the blue band. The algorithm uses reflectance values
from nine bands through each step of the retrieval. After screening and pixel
selection, surface reflectance values are determined based on three bands
(0.412, 0.490 and 0.67
The Cloud Aerosol Lidar with Orthogonal Polarization (CALIOP) on board the CALIPSO satellite has been observing vertically distributed aerosol and cloud properties since 2006. CALIOP is an elastic backscatter lidar operating at two wavelengths (532 and 1064 nm) equipped with a polarization channel at 532 nm (Hunt et al., 2009; Winker et al., 2009). Calibration is achieved through a molecular normalization technique for night-time measurements at 532 nm, which is subsequently the basis for daytime calibrations at both channels (Powell et al., 2009; Kar et al., 2018). The latest CALIOP Version 4 data are significantly improved thanks to the refined calibration algorithms (Getzewich et al., 2018; Kar et al., 2018; Vaughan et al., 2019).
CALIOP data require several processing sequences, handled by different algorithms, to achieve the desired aerosol and cloud properties (Winker et al., 2009). The first algorithm (selective iterative boundary locator – SIBYL) starts by analysing calibrated level 1 data averaged horizontally (at resolutions from 0.33 to 80 km) through the use of an adaptive threshold scheme establishing layer boundary limits (Vaughan et al., 2009). The next steps require the use of scene classification algorithms (SCAs). Firstly, the cloud-aerosol discrimination (CAD) algorithm uses multidimensional probability density functions (PDFs) to distinguish clouds from aerosol layers (Liu et al., 2005, 2009). The primary inputs from the CAD algorithm are later used for subtyping aerosol species throughout the troposphere and stratosphere (Omar et al., 2009; Kim et al., 2018). Finally, SCA uses the attenuated backscatter and volume depolarization ratios (both layer-integrated) to distinguish between water and ice clouds (Hu et al., 2009; Avery et al., 2020). To extract aerosol properties (particulate backscatter and extinction coefficients, optical depth), the classified layer data are fed through several hybrid extinction retrieval algorithms (HERAs) (Young and Vaughan, 2009; Young et al., 2013, 2018).
Lidar ratios are essential for calculating extinction coefficients, and throughout these sequences of algorithms lidar ratios are selected in one of two ways. For unconstrained retrievals, lidar ratios are selected based on the aerosol subtype classification, which is a function of surface type, location, particulate depolarization ratio and integrated attenuated backscatter (Omar et al., 2009; Kim et al., 2018). For each aerosol subtype, a lidar ratio is assigned based on AERONET data, direct measurements and theoretical scattering calculations (Omar et al., 2009; Tackett et al., 2018). The second approach, known as constrained retrievals, is based on measured layer two-way transmittance (Young and Vaughan, 2009; Young et al., 2018). Selection between these two approaches is done based on scene complexity and feature classification (Young and Vaughan, 2009; Young et al., 2018). In most cases aerosol lidar ratios are determined using unconstrained retrievals (e.g. layers in contact with the surface); however constrained solutions are possible in certain situations (Young and Vaughan, 2009; Young et al., 2018).
Figure 1 summarizes the steps of the analysis in detail. Events reported in
scientific literature as well as events that drew significant media
attention within a period of 12 years (2008–2019) were selected, a period
for which both MODIS and CALIPSO were operational. MODIS (aboard Aqua and
Terra satellites) RGB composite images are used to visually identify the
plume. Plumes larger than 500 km
Flowchart of the plume analysis method.
Aerosol properties were only analysed for successful retrievals. The
following aerosol properties were used in our analysis: AOD at 0.55
CALIPSO is used complementarily as it provides important insight into the plume monitoring, being an active sensor. Moreover, CALIPSO flies as part of the A-Train constellation and follows MODIS Aqua observations by 2 min; thus similar atmospheric volume is sampled. The particulate backscatter coefficient (532 nm) is used to define the extent of the plume cross section. Smoke plumes have higher backscatter values than the background aerosol and are easily identifiable in the backscatter profiles. The minimum plume horizontal extent is set to 5 km as this is the standard level 2 data output (Winker, 2018). For daytime, MODIS Aqua RGB images prior to the CALIPSO overpass are used for visual confirmation. Conversely, for night-time, one MODIS image before and one after the CALIPSO overpass are used to assess the plume spatial continuity.
To retrieve detailed information on the aerosol optical properties, we use
CALIPSO Level 2 data – 5 km Aerosol Profile (532 and 1064 nm), standard
version 4.20 (Winker, 2018). The methodology to
quality assure the CALIPSO profiles is mostly similar to the rubric used by
Tackett et al. (2018). For cloud-free scenes, only aerosol profiles with a
cloud-aerosol-discrimination (CAD) score of
In this analysis, the particle backscatter coefficient is used to identify
the geometrical properties of the smoke plume. The plume is defined as the
area where the values are at least 2 times higher than the background, which
is considered an area of identical thickness located either above or
below the plume. The plume AOD (532 and 1064 nm) is calculated by
integrating the particle extinction coefficient in the plume region, and the
plume mean AOD is the average of the individual (i.e. 5 km) plume AODs that
comprise the plume. Additionally, the plume extinction-to-backscatter ratio (i.e.
lidar ratio), Ångström (532
AERONET observations, when available, are also investigated and compared with the satellite measurements. Lastly, in case of events that have already been investigated by means of ground-based or airborne observations, we compared the published results with our methodology, reflecting the impacts oil smoke plumes have on current satellite retrieval capabilities.
Based on the information given in Table 1, we filtered a total of 375 d in which oil smoke plumes were observed by the MODIS sensors. After applying the selection criteria for the MODIS sensor, we obtained a total of 10 d with successful retrievals. The majority of oil plumes resulted in unsuccessful retrievals, 70.7 % while 26.7 % of plumes were screened out due to high percentage of cloud coverage. When applying the selection criteria for CALIPSO, we obtained six plume sections suitable for analysis. Table 2 shows the dates for both MODIS and CALIPSO retrievals suitable for analysis.
List of successful MODIS retrievals and CALIPSO overpass dates.
We selected a successful retrieval to better describe the method used for
our analysis. Figure 2 shows event 14, the case at Ra's Lanuf and As Sidr
tank farms which caught fire on 5 January 2016 and burned
throughout 6 and 7 January. The retrieval in the images was taken
on 6 January by MODIS Aqua at 12:05 UTC. Figure 2a represents a true-colour composite image showing the smoke plumes emerging from both sites and
travelling ENE over the Gulf of Sidra. Judging by this image alone, we can
only distinguish parts of the smoke plume which appear to be less dispersed
and thus present a smaller mixing ratio with the local background aerosols.
In this study, we focused our attention on the plume areas where heavy
concentrations of aerosol are present while discarding retrievals done at
the edges of the plume where background aerosol may have a large influence
on the retrieved values. Thus Fig. 2b was constructed based on the AOD
(0.55
Visual representation of the analysis method for MODIS data:
Figure 3 shows an example of an unsuccessful retrieval of the land algorithm for the event 13 plume on 30 December 2014. We can distinguish the plume from the RGB image over the Gulf of Sidra while also observing AOD values over land where the smoke plume drifted ENE towards the island of Crete. However, there seems to be no distinguishable AOD gradient, over land, in the plume section. A further inspection suggested that all pixels showed values of 0.095, which suggests that the lower radiance values did not match well with pre-existing LUT values. Consequently, the region is classified as “clean atmosphere”, and thus, a unique AOD value is assigned to all the pixels. Conversely, the ocean algorithm retrieved AOD that varied between 0.1 and 0.37. Since these heavy smoke plumes are the result of extreme scenarios, they are rarely observed and may not end up being a subject of research. Thus, we believe there are no cases within the LUT values describing extremely low atmospheric transmission and radiance values, highly absorbent aerosol, low SSA and low reflectance values over a large spectral range including MODIS bands 1 through 7.
Retrieval of plume (unsuccessful) and background AOD values: event
13, 30 December 2014. The red coloured “x” indicates the event origin (satellite
imagery from the NASA Worldview application,
Event 14 at Ra's Lanuf and As Sidr, 6 January 2016, was also
captured by CALIPSO lidar measurements as CALIPSO overpass matched a cross
section of the plume area. Figure 4a shows this overlap in near-real time as
CALIPSO succeeds Aqua within a 2 min time frame. Within the 15 km plume
cross section, we selected a particulate backscatter coefficient profile for
reference, Fig. 4b, and based on this parameter we determined plume
elevation and thickness. The average plume thickness was approximately 920 m. The layer base was situated between 2600 and 3100 m above the Gulf of Sidra while
the top was measured between 3300 and 4200 m. The entire plume cross section
is presented in Fig. 5a. We observe the main plume from Ra's Lanuf
elevated between 2600 and 4200 m. Figure 5a also shows the secondary plume
from As Sidr, 0.2
This event is an example of an opaque aerosol layer, where the lidar did not
penetrate the plume up to the sea surface over the Gulf of Sidra. This event
recorded a lidar ratio of 109
Judging from these images and from the average CAD score of
Cloud formation on top of oil smoke plumes. Upper images depicting
the fire at Balongan, Indonesia, 29 March 2021; lower left image depicting
the fire at Vasylkiv, Ukraine, on 9 June, 2015; lower right image
depicting the fire at Butcher Island, India, on 7 October 2017
(satellite imagery from OpenStreetMap©OpenStreetMap contributors,
2021. Distributed under the Open Data Commons Open Database License (ODbL)
v1.0 and Planet Team,
The current version of the vertical feature mask gives a mixed result for
aerosol typing comprised of dust, polluted dust and smoke aerosols for this
oil smoke plume. The average values for plume AOD ranged between 1.52
The results of the successful MODIS retrievals are presented in Tables 3 and 4 in the form of mean and standard deviation values. The MODIS Aqua
retrieval presented in Sect. 3.1 was found to be in good agreement with
the Terra retrieval. Event 14 showed a larger difference in plume-specific
AOD values between Terra and Aqua retrievals. However this was to be
expected since the fire spread to several oil tanks between the two
retrievals. Based on these results, we identified no large discrepancies
between the two sensors. For plume and plume-specific AOD, the majority of
values fall within the expected uncertainty interval of the retrieval
algorithm,
Mean and standard deviation values of aerosol properties (AOD, AE,
Mean and standard deviation values of aerosol properties (AOD, AE,
Except for event 4, all the plumes exhibit AE values lower than 1. The
larger AE values from event 4 may be attributed to the different fuel type
since the event surrounding SOCAR's platform no. 10 also involved four gas wells (Business-humanrights, 2015). While AE plume values are
generally low, these extremely low values may not be primarily a direct
result of particle size distribution. MODIS uses spectral reflectance
relations to determine AOD and subsequently AE levels. While other types of
aerosols have a varying spectral reflectance signature, heavy concentrated
black carbon exhibits a flat and linear signature that results in low spectral
reflectance values (Johnson et al.,
1991; King, 1992; Pilewskie and Valero, 1992; Soulen et al., 2000). To
further distinguish between these events and the atmospheric background, we
selected the effective radius based on MODIS LUT. For the ocean algorithm,
Following event 14 in Fig. 2, Fig. 7 shows a visual representation of
successful MODIS retrievals from events 13 and 16. We choose to describe in
detail the events from Libya as they are also analysed based on CALIPSO
retrievals. Moreover, the plumes resulting from these events share the same
locations (As Sidr and Ra's Lanuf). Figure 7a shows plume specific AOD
values ranging from 0 to 0.28. Plumes from As Sidr, event 13, are visible in
the first three rows of Fig. 7. This event was captured over multiple days
while the fire engulfed several oil tanks and subsequently injected higher
amounts of aerosols in the region. Depending on the local background levels,
average plume-specific AOD ranged from
Within the 12-year period we identified three events in the Gulf of Sidra,
events 11, 13 and 14. Apart from event 14, previously described in Sect. 3.1, all the remaining CALIPSO retrievals were unconstrained retrievals.
Event 13, at As Sidra on 29 December 2014, was detected by a
CALIPSO night-time overpass which fell approximately 12 h between the
successful MODIS retrievals of 28 and 29 December 2014. The
plume at As Sidr can be observed in Fig. 8a. Off the coast of As Sidr a
distinct feature above the sea surface reaching 650 m in altitude is
observed. Figure 8b shows a particulate backscatter profile where we can
distinguish a plume thickness of approximately 240 m. This event was smaller
in magnitude with respect to event 14 where multiple storage tank fires
contributed to the same plume mass. In this case, CALIPSO overflew much
closer to the tank farm, also resulting in a narrower plume cross section.
The SIBYL algorithm level 2 products were averaged over a larger 20 km area,
as opposed to the 5 km averaging resolution; thus plume values are harder to
distinguish from background aerosol levels. Backscatter and extinction
values can be seen in Table 5. The plume cross section measured
approximately 3 km, as seen in Fig. 8c and d. Consequently, CALIPSO
identified dusty marine aerosols within the Gulf of Sidra region as evident from the
plume lidar ratio of 37
Backscatter and extinction statistics for plume values based on CALIPSO lidar measurements.
Image of event 13
Mean plume values for lidar-specific aerosol properties (PDR – particulate depolarization ratio; lidar ratio) and uncertainty estimates based on CALIPSO measurements.
Mean plume values of aerosol optical properties based on CALIPSO lidar measurements.
Event 11 was captured inland as the plume cross section was identified 170 km south of the Ra's Lanuf tank depot. Figure 8f shows a particulate
backscatter profile through the plume centre, describing a fairly
inhomogeneous mass of smoke particles. The main plume was concentrated
between 500 and 900 m; however lower concentration may have been mixed
with local dust particles all the way up to 1500 m. Figure 8e shows the
extent of the plume as it travelled southwards inland. As was the case of
the previous event, plume lidar ratios were determined by an unconstrained
solution. Thus values of 55
Event 1, at the Qayyarah oil fields in northern Iraq, was captured by CALIPSO
in three distinct cases. In all three cases CALIPSO overpassed within less
than 35 km southwest from the well fires. The plume particulate backscatter
and extinction coefficients ranged from 2 to 5 times higher than local
background values. The plumes were identified within the PBL, and as a
result, the lidar ratios between 44
Based on CALIPSO measurements, the smoke backscatter and extinction
coefficient ranged from 2 to 9 times higher than background levels. In four
out of six cases, particulate depolarization ratio revealed values between
0.11 and 0.15, resembling moderately depolarizing smoke, while larger values
in two cases were mostly due to the presence of dust particles in the local
atmospheric scene. Apart from one case, all lidar ratios were obtained by
unconstrained retrievals as the plume resided in the PBL. The opaque feature
measured high lidar ratios of 109
As discussed in the introduction section, oil smoke plumes have been rarely observed using ground-based remote sensing instruments such as AERONET sun photometers. We used AERONET version 3 direct sun data to assess the presence of oil smoke plumes. Only one study was found in scientific literature (Mather et al., 2007), which measured aerosol properties of the Buncefield plume at two distinct locations. Here we identified the smoke plume, at event 10, resulting from naphtha tank fires in Vasylkiv, Kyiv Oblast, Ukraine, on 9 June 2015. The smoke plume was also captured in RGB images as seen in Fig. 6, lower left image. Figure 9a shows the distinct signature of the oil smoke plume as AOD values increased significantly in all wavelengths. Figure 9c is a good indication of the increasing particle size with respect to the other days observed in MODIS and CALIPSO data as well. Figure 9d shows the daily evolution of AE with values between 0.45 and 0.9 for the time frame in which the plume was observed. Figure 9b shows AOD values rising as the plume was travelling NE over Kiev. The AERONET station in Kiev is situated approximately 35 km NE of the Vasylkiv tank farm. The peak of the plume was detected at 09:45 UTC when the AOD was 0.68 at 500 nm. Unfortunately, no inversion products coinciding with direct sun measurements were available as the Kiev sky was partially cloudy at the time.
AOD and AE smoke plume values at Kiev on 9 June 2015 and monthly values from June 2015.
The results presented in this study show a wide range of values that are
attributed to a multitude of local factors such as background aerosols,
burning rates, weather conditions, fuel type, time of retrieval and local
geography. Other factors can be attributed to the different types of methods
and algorithms used to retrieve aerosol-specific data. MODIS data showed
relatively low values of plume-specific AOD ranging from
Table 8 lists the oil smoke optical properties from different studies that
utilized similar ground-based or airborne measuring techniques. Overall,
the MODIS AOD estimates are very low compared to the reference studies given
in Table 8. It should be mentioned that AOD values from the Gulf war smoke
plumes are larger for the most part due to the magnitude of the event. These
measurements describe super composite plumes resulting from a large number of
well fires and pool fires. An event that more closely resembles the events
in this study was analysed by Mather et al. (2007), who retrieved AOD values
of 0.28 to 0.68
Oil smoke optical properties from ground-based and flight measurements along with the scientific reference.
CALIPSO AOD measurements from event 14 are similar to the upper ranges
measured by Laursen et al. (1992) while the unconstrained retrievals from
this study more closely resemble the lower bounds from Laursen et al. (1992).
The AOD values registered by Ross et al. (1996) fall out of the range of this
current study. This may be a result of plume dimensions as the Ra's Lanuf
plume cross section was much larger and thicker than the plume described in
Ross et al. (1996). In any case, event 14 is expected to measure the largest
AOD values out of all smoke plumes solely based on the magnitude of the
event. Judging by the results seen in Fig. 8a, an aerosol feature
exhibiting such heavy attenuation in the layers directly beneath would most
likely yield higher AOD values. Particulate depolarization ratio for four
out of six cases reflects the values shown by Okada et al. (1992), indicating
that oil smoke particles are moderately depolarizing. Ceolato et al. (2020)
showed PDR of 0.058 for super aggregates. However these values reflect early
combustion particles unaltered by the effects of coating. The effects of
coating on soot particles can result in PDR larger than 0.1 for oil smoke
as suggested by Okada et al. (1992). The same effects are also visible in PDR
of biomass burning evident in Kanngießer
and Kahnert (2018), Haarig et al. (2018), and references therein. The opaque
feature indicated high lidar ratio values (109
In this study, we examined oil smoke plumes derived from 30 major industrial
events within a 12-year period. To our knowledge this is the first study
that utilized a synergetic approach based on satellite remote sensing
techniques. The MODIS ocean algorithm successfully retrieved aerosol
properties in 10 cases, ranging on average from
Not applicable.
CALIPSO data are available at
AM, NA and AO carried out the conceptualization and methodology. AM, ATR and HIS carried out the formal analysis. AM, ATR and HIS provided visuals for the paper. AM, NA and CSB wrote the initial draft. NP, LTD and DN reviewed and edited the initial draft. NA and CSB acquired funding for the current research. AO provided supervision for the PhD students AM and ATR.
The contact author has declared that neither they nor their co-authors have any competing interests.
Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
This work was supported by the project entitled “Development of ACTRIS-UBB infrastructure with the aim of contributing to pan-European research on atmospheric composition and climate change” SMIS CODE 126436, co-financed by the European Union through the Competitiveness Operational Programme 2014–2020.
This work was supported by the project entitled “Strengthening the participation of the ACTRIS-RO consortium in the pan-European research infrastructure ACTRIS” SMIS CODE 107596, co-financed by the European Union through the Competitiveness Operational Programme 2014–2020.
This research has been supported by the Ministerul Cercetării şi Inovării (grant nos. SMIS CODE 126436 and SMIS CODE 107596).
This paper was edited by Evangelos Gerasopoulos and reviewed by two anonymous referees.