The inventory for open waste burning has been built following the IPCC guidelines (chap. 2 and 5 available at https://www.ipcc-nggip.iges.or.jp/public/2006gl/vol5.html, last access: 19 July 2021) for the estimation of greenhouse gases (GHGs): it includes open residential burning and dump waste burning. This inventory does not include the waste burning practices in incinerators or modern combustion systems, which are already accounted for in the industrial sector.

Open waste burning is estimated using the following expression: 2Ei=WB×EFi, where WB is the amount of solid waste that is burned residentially and in uncontrolled dumps, and EFi is the emission factor of pollutant i. EF values for BC and OC are provided by Keita et al. (2018), from Akagi et al. (2011) and from Wiedinmyer et al. (2014) for the other species (NOx, NMVOCs, CO and SO2).

For each African country, the WB amount is estimated following Sect. 5.3.2 of the IPCC Guidelines for National GHG Inventories (Eggelston et al., 2006), which states 3WB=P×MSWp×Pfrac×Bfrac, where P is the national population given by the World Bank database (http://data.worldbank.org/indicator, last access: 2 November 2016) and MSWp​​​​​​​ is the mass of annual per capita waste production taken from Wiedinmyer et al. (2014). The default value of 0.6 recommended by IPCC is used for Bfrac: this is the fraction of waste available to be burned that is actually burned. Pfrac is the fraction of the population assumed to burn some of their waste either near their residence or in uncontrolled dumps. Values for Pfrac used in our calculations are taken from Wiedinmyer et al. (2014). Pfrac is assumed to be based on national income status, urban versus rural population, and waste collection practices. In Africa, Pfrac may be considered to be 100 % following Wiedinmyer et al. (2014). This value is the value obtained for countries with low income, low middle income and upper middle income, following the classification of the World Bank (http://data.worldbank.org/country, last access: 19 July 2021). In this context, data for semi-developed countries are not available. Therefore, an overestimation of waste burning practices may be expected in some countries (e.g., South Africa, Morocco, Egypt). However, currently no data exist to avoid this overestimation.

Note that it is possible to distinguish waste burning emissions which occur near the residence from those in uncontrolled dumps. Such practices are highly different in rural and in urban areas. WB can also be calculated using the following equation: 4WB=WBR+WBU, where WBR and WBU represent waste burning in rural and urban areas, respectively, and are defined by 5WBR=MSWp×Bfrac×Prural×(PfracRes)R+MSWp×Bfrac×Prural×(PfracDump)R,6WBU=MSWp×Bfrac×Purban×(PfracRes)U+MSWp×Bfrac×Prural×(PfracDump)U. In Africa, the rural population is assumed to have no organized waste collection; therefore, in rural areas (PfracRes)R linked to residential burning is assumed to be equal to 100 %, whereas (PfracDump)R linked to uncontrolled dumps is 0 %. In urban areas, (PfracDump)U is country-dependent: for example, the fraction of uncollected waste is 0.77, 0.30, 0.40 and 0.76 for Benin, Côte d'Ivoire, Ghana and Nigeria, respectively (Wiedinmyer et al., 2014). Therefore, in Benin for example (PfracRes)U is assumed to be equal to 77 % and (PfracDump)U is 23 %. Rural (Prural) and urban (Purban) populations are provided by the World Bank database (http://data.worldbank.org/indicator). Rural/urban distinction for estimating WB emissions is an important improvement to our inventory, providing details at the local and regional levels which are missing in global inventories.

The final step in the development of the inventory is the disaggregation of the country-level emission totals to the African gridded domain at 0.1∘ × 0.1 ∘ latitude–longitude resolution using appropriate spatial proxies for each sector. Three types of geographic information system (GIS) datasets at a resolution of 0.1∘ × 0.1∘ are used here for disaggregation: (1) 2010 population density given by the Center for International Earth Science Information Network (CIESIN; Gridded Population of the World Future Estimate: GPWFE), (2) African country road networks based on gridded emission files from EDGAR traffic inventory (Janssens-Maenhout et al., 2011) and (3) African power plant networks given by Africa infrastructure (https://powerafrica.opendataforafrica.org/, last access: 19 July 2021). These spatial allocation proxies are used as follows: (1) road networks are used for road traffic emissions; (2) geographical coordinates of power plants are used for energy production emissions; and (3) the population density grid is used for residential combustion sources, industries and waste burning. In the future, we plan to use gridded rural and urban population densities to better disaggregate waste burning and residential emissions in Africa.

Flaring emissions are taken from the inventory developed by Doumbia et al. (2019) for the years 1994 to 2015 using the following equation: 7EXflaring=GFvolume×XEF×ρ, where EXflaring is the emission rate of a pollutant X (kiloton) and GFvolume is the volume of gas flared in billions of cubic meters (bcm). Yearly averages GFvolume by country for the period 1994–2010 are taken from the National Oceanic and Atmospheric Administration (NOAA) (http://ngdc.noaa.gov/eog/dmsp/, last access: 19 July 2021) DMSP (Defense Meteorological Satellite Program) dataset. No data exist for the years 1990–1993. For the period 2012–2015, estimations of GFvolume are based on Visible Infrared Imaging Radiometer Suite (VIIRS) satellite data (Elvidge et al., 2015) and are spatially distributed based on the DMSP 2011 product. GFvolume for 2011 is estimated from 1994–2010 trends and from the 2012–2015 time series as the 2011 DMSP data cannot be used due to an orbital degradation that led to solar contamination.

XEF is the emission factor (EF) for species X in g kg-1 of fuel burned, and ρ is the density of the fuel gas. Typically, the density of the fuel (natural gas) varies between 0.75 and 1.2 kg m-3 depending on the fraction of heavy hydrocarbons present in the fuel (US Standard Atmosphere, 1976). In this inventory, we assume a gas density of 1.0 kg m-3 for converting the volume of the associated gas to mass (E&P Forum, 1994). EFs for various species (CO, NOx, NMVOCs, SO2, OC and BC) are detailed by Doumbia et al. (2019), and they show a large range of uncertainties. We use here the mean EF value given in the Doumbia et al. (2019) paper.

Spatial disaggregation is achieved by overlapping layers of maps including DMSP nighttime light images, gas flare areas, world maritime boundaries and total gas flare volume per country based on a geographic information system (ArcGIS software). The final layer is gridded at a 0.1∘ × 0.1∘ horizontal resolution.

Figure 1 shows the time series of BC emissions for fossil fuel (FF), biofuel (BF), open waste burning (WB) and flaring sources in Africa from 1990 to 2015. The BC emissions increase by 46 %, 67 % and 43 % for the fossil fuel (FF), biofuel (BF) and open waste burning (WB) sources, respectively, during this time period. This change is mainly due to anthropogenic activity increases linked to population growth. Africa's population has grown by 2.5 % yr-1 over the past 20 years, corresponding to a roughly 64 % increase over the period 1990–2015, which is of the same order of magnitude as the increase in biofuel emissions. Biofuel emissions have increased at a higher rate than other sources due to an increase in low-income population in sub-Saharan Africa, where biomass constitutes about 80 % of the total energy consumption (Ozturk and Bilgili, 2015).

Unlike FF, BF and WB, BC emissions from flaring are decreasing (55 % from 1994 to 2015) due to the actions of the Global Gas Flaring Reduction initiative (GGFR). In 1994, BF, FF, WB and flaring contributions to total anthropogenic BC emissions were roughly of the order of 47 %, 11 %, 36 % and 6 %, whereas in 2015 such values are 45 %, 17 %, 35 % and 2 %, respectively. Increases in the contribution from FF can be explained by increases in the traffic fleet and industrialization.

Figure 2 shows the trends in pollutant emissions in five subregions of Africa (Northern, Eastern, Western, Middle and Southern Africa) for the main sources (fossils fuels, biofuels and waste burning) during the period 1990–2015. The list of countries included in these five regions is given in Table S1 of the Supplement. In Africa, all pollutant emissions are generally increasing over the period 1990–2015 at a growth rate of 95 %, 86 %, 113 %, 112 %, 97 % and 130 % for BC, OC, NOx, CO, SO2 and NMVOCs compared to 1990, respectively. Except for OC, emissions of all pollutants have nearly doubled between 1990 and 2015.

The regional contributions of BC, OC, CO and NMVOCs show similar patterns, with the highest values for Western Africa, Eastern Africa and Northern Africa (except for OC, for which Middle Africa emissions are higher than Northern Africa emissions). Southern Africa emits more SO2 and NOx than all the other African regions, followed by Northern Africa. This is due to their large amount of industrial activities compared to the other regions of Africa.

Analysis of BC emissions by region indicate that the highest emitting region is Western Africa with 0.18 Tg (26 %) in 1990 and 0.39 Tg (29 %) in 2015 followed by Eastern Africa with 0.17 Tg (26 %) in 1990 and 0.33 Tg (25 %) in 2015, Northern Africa with 0.15 Tg (22 %) in 1990 and 0.28 Tg (21 %) in 2015, Southern Africa with 0.10 Tg (14 %) in 1990 and 0.17 Tg (13 %) in 2015, and Middle Africa with 0.08 Tg (12 %) in 1990 and 0.16 Tg (12 %) in 2015. Western Africa's contribution to BC emissions shows the fastest growth (26 % to 29 %) compared to the other regions. The highest rate of increase in BC emissions is also observed in Western Africa (117 %) and the lowest in Southern Africa (70 %) over the period 1990–2015. This could be in part explained by the fact that the population growth rate in Africa is higher in Western Africa with a value of 2.66 % yr-1 and lower in Southern Africa with a value of 1.64 % yr-1, as shown by United Nations (2015). Furthermore, differences in BC emissions between Southern and Northern Africa regions with similar fuel consumption (for fossil fuels, biofuels and solid waste) are partly explained by the emission factors which are higher in Northern Africa, which includes more developing countries (four of seven countries) than in Southern Africa (one of five countries).

In terms of SO2 emissions, the highest emitting region is Southern Africa, which is the most industrialized region in Africa, with emissions of 1.19 Tg (73 %) in 1990 and 2.24 Tg (63 %) in 2015. Southern Africa is followed by Northern Africa, Eastern Africa, Western Africa and Middle Africa, respectively. Over the 1990–2015 period, the highest rate of increase in SO2 emissions occurred in Western Africa (224 %) followed by Middle Africa (71 %), Eastern Africa (56 %), Southern Africa (47 %) and Northern Africa (33 %), respectively. As for BC, the highest rates of increase in SO2 emissions during this period are observed in regions where the population growth rate is the highest, i.e., Western Africa (2.66 %yr-1), Middle Africa (3.10 % yr-1) and Eastern Africa (2.71 % yr-1), whereas the lowest rates are found where the population growth rate is the lowest, i.e., Southern Africa (1.64 % yr-1) and Northern Africa (1.87 % yr-1).

Figure 3 shows the spatial distribution of BC and NOx emissions in 2015 for the fossil fuel, biofuel, waste burning and flaring sources. As given in Table 2, which shows the contribution of each of these sources to the total emissions for BC, OC, NOx, SO2, CO and NMVOCs in 2015, total BC and NOx emissions are 1.35 Tg C and 7.90 Tg NO2, respectively. Emission densities are generally the highest along the Gulf of Guinea and in the east and south of Africa. Table 2 also indicates that for Africa, residential combustion is the major source of carbonaceous particles for BC (40 %) and OC (77 %). It is also the main source of CO and NMVOCs with a contribution of 72 % and 53 %, respectively. Open waste burning is the second most important source of BC, OC and NMVOCs, representing 35 %, 15 % and 22 % of the total, respectively. The energy sector is the major source of SO2 and NOx emissions, which constitute 54 % and 29 % of the total, respectively. For NOx emissions, the energy sector is followed by the traffic sector (26 %), whereas industry is the second largest contributor to SO2 emissions (32 %).

The regional contribution from each sector is presented in Fig. 4. The residential sector contributes to more than 50 % of BC emissions in Eastern and Western Africa and just under 50 % in Middle Africa (Fig. 4a) for the year 2015. In Southern and Northern Africa, it contributes to less than 25 % and 10 % of the total BC emissions, respectively. In these two regions, waste burning is the largest source of BC emissions, with significant contributions from the industry and traffic sources compared to Eastern and Western Africa. Waste burning is the second largest source of BC emissions in Eastern and Western Africa. For NOx, the traffic sector is the largest contributor in Western and Northern Africa, with 30 % and 41 % of the total emissions, respectively (Fig. 4b). In Eastern Africa, the residential sector (32 %) is the largest contributor of NOx emissions, followed by the traffic sector (23 %). In Southern Africa, the two largest contributors to NOx emissions are the energy (52 %) and industry (27 %) sectors, respectively.

Figure 5 shows the sectoral contributions of BC emissions for a few countries in southern Western Africa (SWA) and for South Africa with different predominant sources. As previously mentioned, for countries in SWA, BC emissions are dominated by the residential sector, followed by open solid waste burning and the other sources, whose relative importance differs depending on the country. In Nigeria, industry and flaring BC emissions are much more important than in other countries in SWA. It is also interesting to see that traffic is the largest contributor to BC emissions in Benin, which can be explained by the high number of two-wheeled vehicles, which are more polluting than four-wheel vehicles. In Benin, TW vehicles represent 34 % of road traffic emissions and only 16 % in Côte d'Ivoire for example. In South Africa, waste burning is the predominant sector, contributing 42 % to the total BC emissions (18 % residentially and 24 % in dump locations), followed by the residential (24 %), industry (24 %), energy (13 %), traffic (9 %) and flaring sectors. In Côte d'Ivoire, the residential sector is the most important (58 %), followed by waste burning (26 % with 16 % residentially and 10 % in dump locations), traffic (9 %), other sectors (5 %), industry (1 %) and energy (0.1 %). In South Africa BC emissions from the waste burning sector are higher at the dump than at residences compared to Côte d'Ivoire, where there is less organized waste collection. South Africa is the country emitting the highest amount of SO2, with roughly 62 % of the total African SO2 emissions. Such an important contribution is due to coal-fired power plants with emissions in the range 1000–1500 Gg for the period 1999–2012. These numbers are in agreement with the range given by Pretorius et al. (2015), i.e., 1500–2000 Gg for the same period.

Finally, it is interesting to assess the relative importance of each type of fuel in the sectoral emissions. For BC emissions, emissions related to diesel are 7 times and 8 times higher than gasoline emissions in Côte d'Ivoire and South Africa, respectively. Similarly, wood burning leads to emissions that are about 3 times higher than charcoal burning and making in Cote d'Ivoire. SO2 emissions originate mainly from the use of coal, which constitutes 95 % of the emissions in South Africa (51 % from power plants, 31 % from industry, 4 % from the residential sector and 3 % from other sectors). This demonstrates that actions to substitute diesel by gasoline and wood by charcoal in Côte d'Ivoire would be beneficial to reduce BC emissions. In addition, replacing coal with other fuels such as natural gas would also help reduce BC and SO2 emissions in South Africa.

We first compared the emissions provided by this inventory (DACCIWA inventory) to the inventories developed by Liousse et al. (2014) for the year 2005 and with the emissions from Marais and Wiedinmyer (2016), i.e., the DICE-Africa dataset available for 2006 and 2013. Table 3 summarizes the emissions of pollutants (BC, OC, NOx, SO2, CO and NMVOCs) for the years 2005, 2006 and 2013 for these inventories. Fossil fuel and biofuel emissions from the DACCIWA inventory are slightly lower than those given by Liousse et al. (2014) for BC (0.64 instead of 0.69 Tg), NOx (5.08 instead of 5.80 Tg) and NMVOCs (8.33 instead of 8.60 Tg), lower for SO2 (2.53 instead of 4.02 Tg), and slightly higher for OC (4.96 instead of 3.95 Tg) and CO (64.43 instead of 58.6 Tg). These differences may be explained by the use of more recent fuel consumption data and new emissions factors which are taken from direct measurements. The DACCIWA inventory also includes two major sources not considered in the Liousse et al. (2014) inventory, namely open solid waste burning and flaring sources: including these sources increases the 2005 emissions by 36 %, 33 %, 16 % and 24 % for BC, OC, CO and NMVOCs, respectively, as compared to the Liousse et al. (2014) values.

Table 3 indicates that the DACCIWA emissions of BC, NOx, OC and SO2 are at least 39 % larger than those of Marais and Wiedinmyer (2016), except for NMVOCs and CO emissions which are lower (7 %–25 %). The differences between the DACCIWA datasets and the values reported by Marais and Wiedinmyer (2016) are largely due to the use of different emission factors. A table with the sectoral EFs used in this work and in the Marais and Wiedinmyer (2016) inventories is provided in Table S2. The DICE-Africa inventory uses for example only one EF value for road traffic gathering, motor gasoline and diesel fuel. In the DICE-Africa inventory, much lower BC and OC EFs are used than in this study for the residential source: for example, BC and OC EFs in this study are respectively 5 and 7 times larger than the values used in DICE-Africa.

We also compared our inventory for Africa to emissions from the following global inventories: ECLIPSEv5a (Klimont et al., 2017), EDGARv4.3 (Janssens-Maenhout et al., 2011), CEDS (Hoesly et al., 2018b) and CEDSGBD-MAPS (McDuffie et al., 2020). These global inventories are also developed using a bottom-up methodology, where emissions are calculated as the product of IEA activity data and emission factors for combustion sources including diverse levels of details in technology and emission controls.

Figure 6 shows BC and NOx emission trends from our DACCIWA inventory and the four global inventories mentioned in the previous paragraph. Emission values are indicated in Fig. 6 by colored lines except for ECLIPSEv5a, where values are indicated by dots every 5 years. There are large differences among the emissions for BC, while trends are similar. BC emissions from CEDS and ECLIPSEv5a are both slightly higher than those calculated in this work (12 % on average) over the period 1990–2014. BC emissions from EDGARv4.3 and CEDSGBD-MAPS are, however, much lower, with emissions from the DACCIWA inventory up to 37 % higher on average over the period 1990–2010. NOx emissions from the DACCIWA inventory are of the same order of magnitude as CEDS over the period 1990–2009 (difference of the order of 3 %). This difference becomes larger after 2010, where NOx emissions from the DACCIWA's dataset are about 15 % higher than from CEDS. CEDSGBD-MAPS NOx emissions are lower than the CEDS NOx emissions (on average 23 %) and DACCIWA NOx emissions (on average 27 %). For BC, the NOx emissions from DACCIWA are higher on average by 18 % compared to EDGARv4.3 over the period 1990–2010. The NOx emissions from ECLIPSEv5a are the lowest of all inventories. These differences are due to the use of different fuel and activity datasets in DACCIWA (mainly from UNSTAT) and the two global inventories (IEA), as well as differences in emission factor values used in the different inventories. The EFs used in the DACCIWA inventory are taken from direct measurements at the sources in Africa as described in Keita et al. (2018), whereas the EFs used in global inventories are largely based on measurements taken from other regions such as Europe and the US, which are often not appropriate for Africa.

A comparative analysis of sectoral BC and NOx emissions from the different emission inventories considered in Fig. 6 is performed, which provides further information on the differences between the inventories. For this comparison, we use the year 2010, which is the most recent common year for these inventories. The fugitive sector in CEDS and EDGARv4.3 inventories mainly consists of flaring emissions; in the ECLIPSEv5a and CEDSGBD-MAPS inventory, fugitive emissions are included in the energy sector. Figure 7 shows the relative contribution from each sector to BC and NOx emissions in 2010 for the DACCIWA inventory and the four global inventories. While large disparities exist between the inventories, it is clear that the residential sector is the highest contributor to BC emissions in all the inventories. The main differences in BC emissions are largely due to the residential and waste sectors. The contribution of emissions from the waste sector varies greatly among the inventories (35 %, 7 %, 5 % and 1 % for DACCIWA, CEDS, CEDSGBD-MAPS and EDGARv4.3, respectively). It should be noted that ECLIPSEv5a does not consider a waste sector. The traffic and energy sectors are the largest emitters of NOx for both the CEDS (31 % and 27 %, respectively), CEDSGBD-MAPS (18 % and 38 %, respectively) and EDGARv4.3 (34 % and 32 %, respectively) inventories. In the DACCIWA inventory, the contribution from the energy sector (28 %) is slightly larger than from the traffic sector (25 %).

We also compared OC and SO2 emission trends from our inventory and the four global inventories, as shown in Fig. S1. The DACCIWA inventory shows the highest value for OC emissions. This is mainly due to the value of the OC emission factor used for domestic fires (Keita et al., 2018), which is higher than previous values found in the literature based on measurements from other regions of the world with different wood species. In contrast to OC, SO2 emissions from DACCIWA have the lowest values compared to other inventories. This could be due to differences in activity database and to the EF used for NOx and BC emissions.

Finally, Liousse et al. (2014) have estimated the emissions of BC and OC in 2005 and 2030, using different scenarios. The authors defined the REF scenario as the state of the world for “business and technical change as usual” conditions, driven solely by basic economics. Another scenario has been proposed (called CCC*), where the introduction of carbon penalties and Africa-specific regulations are implemented to achieve a large reduction in emissions from incomplete combustion. These values are shown by the dots in Fig. S2 for the REF and CCC* scenarios. We have linearly extrapolated the DACCIWA emissions for these two species, as shown by the plain lines in Fig. S2. Our estimates for BC and OC are higher than the best-case scenario values and lower than the worst-case scenario values of Liousse et al. (2014). However, OC values are much closer to the worst-case scenario, and BC values are closer to the best-case scenario. These results demonstrate that emission mitigation measures need to be implemented urgently in Africa in order to avoid such elevated emissions in 2030.