Analysis of wildfires and associated emissions during the recent strong ENSO phases in Southern Africa using multi-source remotely-derived products

Abstract In southern Africa, drier conditions are more pronounced during the El Niño Southern Oscillation (ENSO) years, triggering wildfire activity and extreme drought conditions which, individually or together, lead to loss of crop productivity, deaths of livestock and wildlife, famine, degraded ecosystems, water quality and quantity. However, the fire characteristics in relation to the emissions from biomass burning and surface properties are only examined to a limited extent in the literature, especially in Africa, where anthropogenic activities largely determine the fire activity. This study uses the available data from multi-source remote sensing platforms to (1) analyse the spatial distribution of wildfires and associated emissions during strong El Niño (2015/2016) and La Niña (2010/2011) phases in southern Africa, and (2) examine the effects of the severe El Niño and La Niña years on the relationship between the emission parameters, vegetation parameters and climatic parameters. Generally, the results suggest more emissions from the wildfire in the El Niño phase than that of the La Niña. Overall, the Pearson’s correlation clearly shows the influence and the relationship between the climate parameters themselves and also with emission parameters.


Introduction
Wildfires are among the major and prevalent environmental disturbance agents of our time and affect human health, the infrastructure and the earth-atmosphere mechanisms (Zhang et al. 2016). However, the widely studied repercussions of wildfires include burned vegetation scars (De Sales et al. 2019) and aerosols released into the atmosphere during combustion (Wu et al. 2011). Uncontrolled wildfires can lead to serious agroeconomic losses (Bowman and Johnston 2005) and poor air quality (Haywood et al. 2008). Andela and van der Werf (2014) demonstrated that the occurrence and severity of wildfires are particularly widespread in Africa, accounting for nearly 70% of the world's burned area, and these are expected to intensify with warmer and drier climatic conditions (Abatzoglou 2016). In southern Africa in particular, the annual burned area is projected to increase by 5.4% (Andela and van der Werf 2014). The arid and semi-arid regions are more susceptible to wildfire activity (Gaveau et al. 2013), and these conditions are typical of southern Africa with large fluctuations in rainfall and regular droughts. In this region, rainfall fluctuation is strongly regulated by the El Niño Southern Oscillation (ENSO) phenomenon via teleconnections (Anyamba et al. 2002).
Several previous studies have associated the increase in wildfire activity during drought conditions to ENSO (Swetnam and Betancourt 1998;Westerling et al. 2006;Verhegghen et al. 2016;da Silva et al. 2018). The relationship is such that severe El Niño conditions affect rainfall patterns and induce drier vegetation conditions over southern Africa (Mennis 2005), increasing the vegetation's susceptibility to wildfire (Nepstad et al. 1999). For example, several regions such as the Amazon Forest of Brazil (Nepstad et al. 1999), the province of East Kalimantan in Indonesia (Nugroho 2006), central, south and west Kalimantan in Indonesia (Purnomo et al. 2021), Canada (Skinner et al. 2006), Swaziland (Dlamini 2007), and more recently also in the Republic of Congo (Verhegghen et al. 2016) endured widespread fires during strong El Niño periods. On the other hand, Nicholson and Selato (2000) showed that La Niña appears to have the greatest influence on rainfall in southern Africa and wet episodes tend to occur throughout the subcontinent during the first few months of the post-La Nina year. Furthermore, Anyamba et al. (2018) studied the precipitation anomalies over Southern and Eastern Africa during the 2015-2017 ENSO cycle and found that countries in Southern Africa experienced dry conditions during El Niño and wet conditions during the La Niña phase. Despite the meteorological influence, ENSO has an impact on emissions from wildfires. However, the response of wildfire emissions to various ENSO cycles is rarely studied in the literature.
Accurate and regular fire observation in near-real-time and with suitable spatial scales is an indispensable resource for improving our understanding of fire behavior, which significantly enriches logistical fire monitoring and planning strategies (Flannigan et al. 2009). With fires spreading across large and even remote locations, satellite remote sensing is now a practical tool for assessing the disturbance of vegetation ecosystems by fires and the associated effects (Fernandez-Manso et al. 2019). Compared to field observations, remote sensing images offer an objective, flexible and inexpensive alternative to strategically and consistently capture burned areas over expansive areas (Llover ıa et al. 2016). For example, Moderate Resolution Imaging Spectroradiometer (MODIS) has been used as the preferred choice for assessing burned areas in many studies since its inception two decades ago (Freeborn et al. 2011;Cheng et al. 2013). Moreover, data from satellites such as the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) and Atmospheric Infrared Sounder (AIRS) have been proven useful for quantifying and characterizing regional atmospheric emissions from various sources and prevailing meteorological conditions (Freeborn et al. 2016).
Against this backdrop, the current study explores the responses of wildfires and associated emissions to recent strong ENSO phases, i.e., 2010ENSO phases, i.e., -2011ENSO phases, i.e., and 2015ENSO phases, i.e., -2016ing the strong La Niña and El Niño phases, respectively. Specifically, the study examines (1) the spatio-temporal of characteristics wildfires during the recent strong La Niña and El Niño phases in terms of emissions and burned area distribution, and meteorological parameters using multiple data sources, and (2) assesses the relationship between the emissions, meteorological and vegetation parameters during these ENSO phases. The study was conducted over southern African region, characterized by different climatic regions and vegetation types.

Study area
This study was conducted over the Southern African region (Figure 1). From April to October, the area generally experiences dry conditions, therefore the fire occurrence of fires coincides with this period and peaks in August (Zubkova et al., 2019). This region is a largely arid to semi-arid region, characterised by various land covers, topography, and bioregions. Most of the study area comprises temperate Grasslands, semi-arid Savanna, Grassland and Shrubland, Mediterranean forests and cropping activities.

Data and methods
The data and methods for this study are summarized in Figure 2.

Fire emissions
When biomass is burned, a combination of air pollutants are injected into the atmosphere in the form of particulate matter and gases. These harmful constituents affect the climate system, human health and the environment. The fire emission parameters used here are from the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2), CALIPSO and AIRS. MERRA-2 is a NASA atmospheric reanalysis data from 1980s and replaces the existing MERRA reanalysis with an updated product of the Goddard Earth Observing System Model, Version 5 (GEOS-5) data assimilation system. The model includes the dynamic finite volume core which uses a cubed-sphere horizontal discretization at an approximate resolution of 0.5 Â0.625 and 72 hybrid-eta levels from the surface to 0.01 hPa. More details on MERRA-2 can be found in Gelaro et al. (2017). In this study, the monthly parameters, i.e., black carbon (BC) surface concentration (mg m À3 ), BC Biomass burning emissions (mg m À3 s À1 ) and sulphur dioxide (SO 2 ) Biomass burning emissions (mg m -3 s -1 ), at the spatial resolution of 0.5 Â0.625 are retrieved from MERRA-2. The Elevated smoke aerosol optical depth (AOD) at 532 nm (Mm À1 ) and smoke extinction coefficient are retrieved from CALIPSO. The specifications of CALIPSO are discussed in detail by Winker et al. (2003) and(2010). From AIRS, we use carbon dioxide (CO 2 ) and carbon monoxide (CO) at the spatial resolution of 13.5 km at nadir and 1 km vertical resolution. We refer the interested readers to Morse et al. (1999) and Lambrigtsen et al. (2004) for further information on AIRS.

Wildfire characteristics
The wildfire characteristics were assessed using the MODIS-derived Burned Area (BA) (MCD64A1) and MODIS Fire Location (MCD14ML) products available from NASA Earth data (https://lpdaac.usgs.gov/products/mcd64a1v006/) and Fire Information for Resource Management System (FIRMS, https://firms.modaps.eosdis.nasa.gov/), respectively. The MCD64A1 is a global monthly burned area dataset with a ground sampling distance of 500 m, with each pixel representing the day-of-burn as detected during the MODIS satellite overpass. The original burn date Collection was aggregated for each El Niño and La Niña period in Google Earth Engine (GEE), to determine the dates of burn during these periods. The fire locations are calculated using the MCD14ML c6 algorithm, which categorizes actively burning fires during overpass using MODIS thermal infrared (i.e., 4 mm, 11 mm and 12 mm) and reflectance regions (i.e., 0.65 mm, 0.86 mm, and 2.1 mm) (Giglio et al. 2006(Giglio et al. , 2018.

Meteorological parameters
Meteorology has a strong influence on wildfires. The most important meteorological parameters that help generate extreme fire behaviour include low relative humidity, strong surface winds, unstable air, and low precipitation. Meteorological parameters used in this study were retrieved from various sources, namely, the AIRS, Tropical Rainfall Measuring Mission (TRMM), AIRS and CALIPSO. AIRS also measures (i.e., in addition to above mentioned greenhouse gases) meteorological parameters such as air temperature (AT), water vapour and relative humidity (RH) at the same resolution as greenhouse gases (see Section 3.1). The monthly RH (%) and air temperature ( C) meteorological parameters are used in this study.
TRMM, on the other hand, provides precipitation records through a variety of spacebased tools to improve our knowledge of the relationship between clouds, precipitation (PCPN) and water vapour at the spatial resolution of 0.25 Â0.25 , which are critical to the Earth's climate. Further information on the TRMM's product is provided by Simpson et al. (1988) and Kozu et al. (2001). The monthly precipitation rate data (mm/month) is utilized in this study. The spatially distributed monthly temperature data at the spatial resolution of 2 Â5 is retrieved from the Cloud-Aerosol Lidar with Orthogonal Polarization (CALIOP), which is one of the payloads on CALIPSO. Further technical details on the CALIPSO can be found in Winker et al. (2010). Another parameter that is an essential climatic indicator for forest fires is the vapor pressure deficit (VPD). VPD is related to maximum temperatures and low humidity, representing the stress conditions for vegetation prone to fires, however, this parameter is not discussed in this work. More details on VPD can be found in Silvestrini et al. (2011) andFarf an et al. (2021).

Data and statistical analysis
As shown in above (Section 3.2), each parameter has different spatial resolutions, with the parameters from MODIS products (i.e., burned area and NDVI) being the highest (i.e., 500 m -1 km) and emissions parameters from CALIPSO being the worst (i.e., 2 Â 5 ). However, CALIPSO products have been used extensively in previous studies at regional to global scales and provide reasonably accurate measurements of the atmospheric pollutants at various heights. Moreover, sub-Saharan Africa has a lack of ground monitoring stations to avert data shortages. For the current analysis, the differences in the spatial resolution were not a concern since each parameter is analyzed separately in its native resolution. However, we have ensured that all parameters had a common spatial reference system, i.e., World Geodetic System 1984 (WGS-84).
Density analysis was performed in QGIS ver. 3.4 on Fire location data to obtain the fire density maps of the two ENSO phases.
Since the parameters were provided for each month, we aggregated them to obtain the total for the periods under study, i.e., 2010/11 and 2015/16. Using these aggregated parameters, we computed spatial absolute differences in emissions between the El Niño and La Niña phases, to determine if either period yielded higher or lower emissions, respectively.
where x ij is the aggregated emission parameter x (e.g., BC) during La Niña i phase and El Niño j phase, respectively. For further interrogation of the emission parameters, we aggregated each parameter by season.
Pearson correlation (r, Equation 2) analysis was executed to establish the relationship between the wildfires, emissions, meteorological and vegetation conditions. Each parameter statistic (i.e., regional mean) was extracted using the bounding box shown in the insert map of Figure 1, stored in a comma-separated values (.csv) format and subject to r analysis in R-Statistical software ver. 4.1.2 using the rcorr() function from Hmisc R-package, which provides r and p values at 0.05 significance level, while plotting was achieved with corrplot() function from corrplot R-package. The specific parameters tested for their relationship are black carbon (BC), carbon monoxide (CO), burned area (BA), normalized difference vegetation index (NDVI), precipitation (PCPN), air temperature (AT), relative humidity (RH).
where r is the correlation coefficient, variables x i and y i represent values for each respective variable x and y, respectively, and x and y are mean values of the x and y variables, respectively. Further information on the Pearson's correlation test can be found in Benesty et al. (2009).

Comparison of the distributions of emissions during strong ENSO phases
The two phases of the El Niño Southern Oscillations (ENSO), i.e., La Niña and El Niño, are associated with wet and dry conditions, respectively, in southern and eastern Africa. We compared the spatial distributions of wildfire emissions during the 2010/11 and 2015/ 16 periods, where strong La Niña (i.e., sea surface temperature anomaly, SST anomaly <1.5 C) and El Niño (SST anomaly >1.5 C) phases were observed, respectively. The historical sea surface temperature (SST) anomaly indicating weak (i.e., ±1 C-1.5 C), mild (i.e., ±0.5 C-1 C) and strong (> ±1.5 C) La Niña and El Niño phases are displayed in Figure 2. The recent strong ENSO phases are the focus of the current study and are observed when the SST anomaly in the Niño 3.4 region (5 N-5 S, 170 W-120 W) rises above ±1.5 C (NOAA Climate Prediction Center). These phases occurred in 2010/ 2011 (i.e., strong La Niña phase) and 2015/2016 (i.e., strong El Niño phase) and are shown in grey-shaded areas in Figure 3. It is worth noting that there is some slight overlap in the La Niña period with some El Niño conditions in the beginning of 2010. This overlap impacts the transitioning of the climatic conditions from El Niño to La Niña conditions. Nonetheless, we hypothesize that the La Niña and El Niño conditions play different yet crucial roles in the occurrence, spread and intensity of wildfires. Since El Niño is associated with drier and hotter conditions, it favors the occurrence and spread of wildfires in Southern Africa. In contrast, the wet and colder conditions provided by La Niña discourage the proliferation of wildfires.
The period 2013/14, considered a neutral (i.e., non-ENSO) period, was used as a benchmark to highlight the differences in emissions between the two ENSO periods. The results (Figure 4) show the spatial distribution of BC surface concentration, BC biomass burning emissions and SO 2 biomass burning emissions during the neutral period, strong El Niño and La Niña phases. The BC surface concentration comprises the total BC emissions from various sources such as biomass burning, motor vehicles, industrial processes, and household energy, to name a few. BC biomass burning emissions, on the other hand, is strictly emissions from the burning of organic materials such as grasses and trees. Figures 4a-c shows the highest BC surface concentration of $2.5 mg m À3 in the southwest of the Democratic Republic of Congo (DRC) and spreads at moderate values (defined as a range between 0.7 and 1.4 mg m À3 ) towards the west in Equatorial Guinea and northern Angola all located at a region between 0 and 10 S. The moderate BC surface concentration of $1 mg m À3 is also observed in the central and southeastern parts of South Africa and also parts of Botswana, Namibia and Southern Angola at a region between 10 and 30 S. The comparison of between the two ENSO periods under study using Absolute Difference in emissions shows that El Niño phase has higher amounts of BC surface concentration (i.e., >0.5 mg m À3 ) than the La Niña phase in the DRC (see Figure 4d); while the La Niña phase has higher BC surface concentration (i.e., $0.3 mg m À3 ) than El Niño phase over countries such as Botswana and parts of northern Namibia and Southern Angola.
The results for BC biomass burning emissions during the neutral period (Figure 4e), El Niño ( Figure 4f) and La Niña (Figure 4g) phases, show the BC biomass burning emissions of $0.045 mg m À3 during both ENSO phases in the areas which are dominantly covered by forests. The comparison of the two ENSO phases using the Difference show that La Niña phase is characterized by a significantly higher BC biomass burning emissions of $0.02 mg m À3 than El Niño phase in the Kalahari Desert region (i.e., Botswana), northern Namibia and southern Anglo (Figure 4h), which are predominantly grassland and shrubland (see Figure 1). In contrast, the results also show the sparsely distributed higher BC biomass burning emissions over the DRC during the El Niño phase than the La Niña phase. The results for SO 2 biomass burning emissions (Figure 4i, 4j and 4k) generally exhibit the similar spatial patterns as BC biomass burning emissions. The comparison SO 2 biomass burning emissions between the two ENSO phases using the Difference (Figure 4l) shows a relatively higher emissions (i.e., þ0.02 mg m À3 ) in the eastern RSA during La Niña phase when compared to El Niño phase. However, a noticeable difference is the SO 2 biomass burning emissions hotspot in the southeast of RSA.
Another critical emission from wildfires is smoke aerosols, which is predominately from forest burning, as well as other minor contributors such as motor vehicle emissions, industrial activities, and domestic fires, to name a few. Here, we compared the spatial and vertical distributions of Elevated smoke AOD and smoke extinction coefficient during the two recent strong ENSO phases (Figure 3). The results of the spatial distribution of smoke AOD (Figure 5c) show sparsely distributed patches of smoke which are relatively higher during the El Niño phase. Conversely, parts of RSA (i.e., central and southeastern RSA) as well as parts of southern Kenya and Uganda, exhibit relatively higher smoke AOD during the La Niña phase. Figure 5d and 5e shows the latitude-height distribution of the smoke extinction coefficient during the La Niña and El Niño phases. Unlike the spatial distributions, the vertical distributions of smoke show distinct profiles between the La Niña and El Niño phases. The latitude-height distribution profile is presented in different regimes, i.e., the red blocks, in Figure 5d and 5e. During the La Niña phase (see Figure 5d) three smoke emissions regimes are observed namely, A (35 S À 25 S), B(18 S À 9 S) and C(5 S À 0 S). In Figure 5d, regime A shows low smoke extinction coefficients <1 Mm À1 at altitudes of 3 km and 6 km. This is smoke from wildfires in the southern RSA, where accidental fires in forest plantations are common. Moreover, the sugarcane belt of RSA and grasslands used for livestock products are within this region which are dominated by the use fire after harvesting and grazing management activities, respectively. Regime B also shows low smoke extinction coefficients <1 Mm À1 but at altitudes of between 2 km and 8 km. As Figure 5a depicts, this smoke is from the wildfires in the southern DRC and northern Zambia and Angola, which are all dominated by sub-tropical forests. Regime C shows low to moderate smoke extinction coefficients between 0.5 and 1 Mm À1 at altitudes between 1 km and 5 km, which can be attributed to the forests clearing activities in the DRC.
During the El Niño period (see Figure 5e) two smoke emission regimes are observed namely, A (35 S À 18 S) and B (15 S À 5 S). Regime A, which corresponds to RSA, Zimbabwe, and Mozambique smoke spatial distribution (see Figure 5b), shows low smoke extinction coefficients <1 Mm À1 at altitudes of between 1 km and 9 km. On the other hand, regime B-which corresponds to the smoke distributions in the Northern Angola, Northern Zambia, Malawi, Southern Tanzania and the southern DRC-shows low to high smoke extinction coefficients between 0.5 and 2.5 Mm À1 at altitudes between 2 km and 6 km. The mean smoke extinction coefficient profiles during the La Niña and El Niño phases are shown in Figure 5f and 5g, respectively. The highest value of smoke extinction coefficient of $7.8 Mm À1 at an altitude of 2 km is observed during the El Niño phase. However, a thick plume of smoke, i.e., $2 km, is observed between 2 and 4 km during the La Niña phase. Pockets of smoke parcels are also observed at 1 km and 6 km during both ENSO phases. El Niño phase, on the other hand, has lots of peaks indicating smoke from the surface to $9 km.

Spatial distribution of wildfires during strong ENSO phases
The spatial patterns of wildfires, characterized by the burned area (BA) and the fire density, during the strong La Niña and El Niño phases in 2010/11 and 2015/16, respectively. As shown, the spatial distribution of BA is significantly higher during La Niña than El Niño, with Botswana, Northern Namibia, Angola, Zambia, Mozambique, and Southern DRC exhibiting large fire scars (see Figure 6a and 6b). On the other hand, the other countries in the region exhibit more or less the same burned area during the two periods. Moreover, they exhibit high fire density (i.e., 2000 to 2500 fires per 50 km 2 ), while the southernmost countries such as Botswana, Namibia, Zimbabwe, and South Africa and the latitudes around 0 <5 exhibit relatively low fire density, i.e., 0 À 500 fires per 50 km 2 , in both periods.

Meteorological conditions and their relationship with wildfire emissions
The observation of the meteorological parameters is key to understanding the burning characteristics and emissions from wildfires. The AT observations from AIRS (Figure 7a and 7b) during the two ENSO phases show generally higher mean AT of $70 C across most of the southern Africa during El Niño phase, while the mean AT during La Niña are <50 C for most parts, with exception of the DRC where the AT is high, i.e., $55 C. The absolute difference results (Figure 7c) indicate a significantly higher AT by up to $15 C in the most of the Southern Africa during the El Niño phase, while most DRC, northern Angola, and coastal Mozambique and Madagascar show relatively higher temperatures during the La Niña. Relative humidity (RH) during the two ENSO phases (Figure 7d and 7e) show moderate RH of 60% in the DRC and a relatively low RH of between 20 and 35% in countries south of the DRC. The RH comparison using absolute difference (Figure 7f) shows higher RH of up to 8% in the countries south of the DRC during El Niño phase. The DRC itself does not show any major changes in RH in the El Niño and La Niña phases. The precipitation (PCPN) rate results (Figure 7g and 7h) show that during the El Niño and La Niña phases. A moderate precipitation rate of $200 mm/month is observed in the DRC for both ENSO phases. This indicates that this area receives a sufficient amount of rainfall annually. Countries south of the DRC receive a relatively higher PCPN rate of less than 100 mm/month. The absolute difference in PCPN rates (Figure 7i) shows that countries south of the DRC exhibit a slightly higher precipitation rate of between 20 and 60 mm/month during the El Niño phase, while Northern Angola, parts of Madagascar, Tanzania, and Mozambique show relatively higher PCPN rate during La Niña phase. The DRC, on the other hand, shows a combination of high, low and no change in the PCPN rate between the two ENSO phases distributed across the country.
Next, we evaluated the relationship between the meteorological parameters, vegetation conditions and emission parameters using the Pearson's correlation coefficient (r). The results of the statistical relationships between different parameters during the two ENSO phases are presented in Figure 8a and 8b. For ease of interpretations, the r coefficients were grouped into negligible (±0.0 to ±0.3), weak (±0.31 to ±0.5), moderate (±0.51 to ±0.7), high (±0.71 to ±0.9), and very high (±0.91 to ±1.0) negative or positive correlations. The crossed-out r coefficients in Figure 10 indicate statistically not significant correlations at a ¼ 0.05, hence are not discussed. The results indicate that during La Niña, the burned area (BA) shows a moderate positive correlations with air temperature (AT), RH, and precipitation rate (PCPN). On the contrary, during El Niño, a moderate positive r coefficient is observed for AT only, while the RH and PCPN show weak positive correlation. The results imply that BA tends to increase with increasing AT, RH, and PCPN. The PCPN causes an increase in the fuel availability (i.e., vegetation biomass) as the high positive r coefficient confirms the relationship between PCPN and NDVI, while higher AT causes the vegetation to dry, thus resulting in increased BA. A weak correction between BA with RH and PCPN Figure 7. Mean spatial distribution of temperature (a-c), relative humidity (d-f) and precipitation (g-i) during the La Niña and El Niño phases.
is because these parameters are very low during the El Niño periods, while BA is relatively higher due to dry conditions. It should be noted that these are not the only factors affecting BA. Previous related studies (Kganyago et al. 2021) found that wind speed and Palmer Drought Index had moderate positive correlations of 0.68 and 0.52, respectively, with BA.
Another interesting observation for both La Niña and El Niño phases is the moderate to strong negative correlations of BC and CO with AT, RH, PCPN and NDVI, which are all statistically significant at a ¼ 0.05. This indicates that both emission parameters tend to increase with declining the meteorological and vegetation conditions and vice versa. This observation is reasonable since, for example, higher PCPN and NDVI (i.e., vigorous green vegetation) is associated with lower wildfire occurrence and emissions, while the converse would favor the spread of wildfires and cause high BC emissions. Overall, the Pearson's r correlation results clearly show that the meteorological parameters, i.e., AT, RH, PCPN, vegetation conditions, i.e., NDVI, and emissions, i.e., BC and CO are strongly negatively correlated in both ENSO phases, while BA is positively correlated to AT, RH, and PCPN. BA is not significantly correlated to emission parameters studied here.

Discussion
The influences of the El Niño Southern Oscillation (ENSO) phenomenon has been of interest for decades due to its strong influence to climatic phenomena of various regions. Of interest, the extreme conditions introduced by the different ENSO phases, i.e., La Niña and El Niño, such as floods and droughts, respectively, have cascading effects on the ecosystems, air quality, and human livelihoods. However, studies focusing on these cascading effects in an African context such as the effects of ENSO on the wildfires regimes and associated biomass burning emissions, are rare. As a result, the current study sought to address this gap by assessing the responses of wildfires and associated emissions to the recent strong ENSO phases, i.e., 2010/11 and 2015/16. Generally, the results showed that both strong ENSO phases, i.e., 2010/11 (La Niña) and 2015/16 (El Niño) have a considerable influence on the emissions in southern Africa, with different areas exhibiting different emission regimes. The results showed that conditions caused by the 2015/16 El Niño phase resulted in moderate to high BC surface concentration in the region between 0 and 10 S. The drier conditions and associated climatic phenomena such as droughts during the El Niño phase, seem to affect the tropical forests which characterizes most of this region, thus favoring the ignition of wildfires and their spread. In particular, the 2015/16 El Niño phase has been considered the worst ever recorded with sea surface temperature (SST) anomaly of >2 C, thus the observed higher emissions are due to high occurrence of wildfires due to the favorable conditions (such as dry vegetation biomass and high temperatures). Moreover, forests have many components that can burn such as leaves, branches, and tree trunks, as well as the understory herbaceous vegetation. Therefore, the burning of the forests releases large amounts of BC aerosols into the atmosphere due to their high fuel load. Under dry conditions and sustained higher AT (as observed in Figure 7c), understory fires may be easily ignited by lightning or escaped fires from agricultural fires. With a rich biodiversity of fauna and flora, forests in most African countries are often close to human corridors and provide various ecosystems services. However, fires are often used for hunting and clearing the forest land in favor of agricultural expansion, thus may cause widespread forest fires as observed in Figure 8b over Angola, Zambia and DRC. In contrast, La Niña phase brings cold and wet conditions which can wash down and reduces the BC aerosols from the atmosphere during rainfall (Chen et al. 2016), thus slightly decreasing BC concentration. This effect was evident in the DRC, where the relatively lower BC concentration ( Figure  4b) and higher PCPN (Figure 7i) were observed.
Conversely, conditions introduced by the La Niña phase resulted in higher BC surface concentration of $0.3 mg m À3 in the region below 10 (see Figure 3). This region also experienced greater BA during the La Niña phase (see Figure 6a). Moreover, countries such as Namibia and Botswana are characterized by several desserts such as Kalahari and Namib deserts. Therefore, as shown by Rosenfield et al. (2001), desert dust can inhibit PCPN, thus these areas experience lower rainfall even during La Niña that only has slight or no effect on the BC surface concentration. Moreover, the desert areas and semi-arid southern Africa are comprised of grasslands and shrublands, which are well adapted to fire and drought (Ratajczak et al. 2014) and they generally do not release large amounts of BC aerosols during burning. This explains larger BA during La Niña phase which is comparable to the El Niño phase. Although the results have shown that wildfire patterns (i.e., BA and fire density) between both ENSO phases were generally similar, the BA was widespread during La Niña (see Figure 6). This observation suggests that there are other factors (such as human activities) that influence wildfires in southern Africa while El Niño and La Niña only make it more severe in various pollutant indicators, this is supported by Figure 9. An increase in CO emission is observed when taking the difference between La Nina-normal year in most parts of Southern Africa such as Zimbabwe and Angola regions (see Figure 9a). Inversely, the difference in the El Nino-normal year sees a decrease in CO emissions in some parts of Southern Africa (see Figure 9b). Furthermore, the difference in La Nina-normal year for BC biomass burning emissions shows major increases in Zimbabwe and some parts of Angola and Namibia (see Figure 9c). However, decreases in BC biomass burning emissions are present in some parts of Southern Africa such as the DRC. The difference in El Nino-normal year (see Figure 9d) shows a complex pattern of increases and decreases in BC biomass burning emissions.
It should be noted that the years preceding 2010/2011 (i.e., La Niña phase) were characterized by a strong to moderate El Niño in 2009/2010 (see Figure 2), which was dissipating the beginning of 2010. Therefore, the hot and drier conditions, characteristic of the prior El Niño phase, caused widespread wildfires and emissions which were eminent even during the following La Niña phase (Note that this result is only based on the latest event). This prior El Niño phase increased the BC and SO 2 aerosol loading as a result of lingering drought conditions at the beginning of 2010. Certainly, the effects of the dry or moist conditions from each ENSO period may linger until the vegetation conditions improve due to sustained La Niña conditions or until vegetation dries up due to sustained El Niño conditions, respectively. Generally, it is anticipated that, during La Niña, heavy rainfalls will increase the vegetation vigor and wet the fuel (i.e., dead vegetation biomass) making it difficult to burn, thus reducing the emissions of BC and SO 2 aerosols. However, as observed here, this was only true for the DRC, where La Niña phase has relatively lower emissions than El Niño phase, which similarly exhibits lower emission than La Niña phase at the region south of the 10 latitude.
Another interesting observation is the hotspot in the southeastern RSA (see Figures  4i-k), which is characterized by sugarcane farming. This area is known as the sugarcane belt of RSA and the area is burnt seasonally , i.e., exhibiting 5001 À 1000 fires/50 km 2 (see Figure 6c). Controlled fires are the main and cheap agricultural management strategy used for land clearing after harvesting and sugar processing . This causes, among other constituents, releases of BC and SO 2 biomass emissions. However, the release of these emissions seems to be influenced by phelogical crop development, grazing management cycles (i.e., every two years for high biomass) and agricultural land management decisions. The comparison SO 2 biomass burning emissions between the two ENSO phases also showed a relatively higher emissions by 0.02 mg m À3 (see Figure 4l) in the eastern RSA during La Niña phase, which further solidifies that relatively high PCPN expected during La Niña phase did not influence the fire regimes and associated emissions. Consistently, Shikwambana et al. (2021) showed that there is an increasing trend of SO 2 from sugarcane since the 1980s. Moreover, the Pearson's r correlation analysis showed similarly strong negative correlations at a ¼ 0.05 between biomass emissions (i.e., BC and CO) and agro-meteorological parameters (i.e., AT, RH, PCPN and NDVI) during both ENSO phases (see Figure 8). The finding implies seasonal effects as the agro-meteorological conditions would naturally decline off-season, causing expansive wildfires and increase during the growing season, thus decreasing the chances of wildfires. Similar studies have been conducted by Wang and Li (2022), Chao and Min (2022) and Owoade et al. (2012).
It is, therefore, apparent that the ENSO phases did not influence the wildfire regimes and emissions in primarily agricultural areas as they are mainly influenced by rainy seasons and farming practices. It may be beneficial to explore the influence of the two strong ENSO phases on wildfire emissions by season and land cover in future studies, so that the seasonal and land management influences can be decoupled. Moreover, analysis over longer periods of time and incorporating comprehensive list of relevant parameters is recommended for future studies.

Conclusion
ENSO is a climate-forcing mechanism that has been shown to affect precipitation and the occurrence of wildfires in many parts of the world. To our knowledge, only a limited number of studies of this kind have been conducted in southern Africa. This research focused particularly (1) on the spatial distributions of emissions, burned area and meteorological parameters, and (2) the relationship between the emissions parameters, agro-meteorological parameters under the strong La Niña and El Niño phases. Overall, the results show that all the pollutants under study are impacted by both ENSO phases. However, the magnitudes of the pollutants differ slightly for each phase. In general, a magnitude of pollutants was found to be region, season and land cover dependent. For example, region below 10 S exhibited relatively higher biomass burning emissions during the El Niño phase, while the region above 10 S exhibited relatively higher emissions during La Niña phase. Regions characterized by Forests and Croplands released the most BC emissions, while areas characterized by grasses and shrubs caused relatively lower BC and SO 2 biomass burning emissions. Seasonal effects need to be decoupled in future studies. Furthermore, in the La Niña phase precipitation showed a strong negative correlation with BC and CO, and showed a positive strong correction between BA, AT and RH. El Niño phase showed a similar correlation except for BA where the correction was weak. A weak correction between BA and BC is a surprise as it was anticipated that there would be a direct relationship between these parameters i.e., large BA would result in large BC emissions. Another surprising result is the low correlation between BA and NDVI. Again, it is anticipated that a low NDVI would indicate drier vegetation which is easier to burn, thus leading to a large BA. Overall, this study has provided new insights into the relationship between these parameters and will be extended to other regions.