Emission inventory processing of biomass burning from a global dataset for air quality modeling

Wildfires generate large amounts of atmospheric pollutants yearly. The development of an emission inventory for this activity is a challenge today, mainly to perform the air quality modeling. There are accessible available databases with historical information about this source. The main goal of this study was to process the results of biomass burning emissions for the year 2014 from the Global Fire Assimilation System (GFAS). The pollutants studied were black carbon, organic carbon, and fine and coarse particulate matter. The inputs were pre-formatted to enter into the simulation software of the emission inventory. In this case, the Sparse Matrix Operator Kernel Emissions (SMOKE) was used, and the values obtained in various cities were analyzed. As a result, the spatial distribution of the forest fire emissions in the Southern Hemisphere was achieved, with the polar stereographic projection. The highest emissions were located in the African continent, followed by the northern region of Australia. Future air quality modeling at a local level could apply the results and the methodology of this study. The biomass burning emissions could add a better performance of the results and more knowledge on the effect of this source


Introduction
Biomass burning, also known as vegetation burning, is considered a significant emission source of atmospheric pollutants and includes wildfires, controlled agricultural burns, and biofuels (Levine 2003). It is estimated that anthropogenic-started fires account for 90% of all wildfires, being the difference those natural fires triggered by atmospheric lightning (Levine 1991). Several studies have been reported on the effect of the biomass burning emissions in climate (Thornhill et al. 2018), photochemistry of the atmosphere (Yue and Unger 2018), biogeochemical cycles (Chen et al. 2010), and human health (Apte et al. 2018).
The estimation of an air emission inventory from wildfires is complex due to the large spatial and temporal variability of this source at a regional scale (Kaiser et al. 2006;Andreae 2019). In order to bring more accurate information, some systems like Moderate Resolution Imaging Spectroradiometer (MODIS) and National Polar-orbiting Operational Environmental Satellite System (NPOESS) monitor and forecast air quality from satellite-based observations of the burnt area (Reid et al. 2009). Therefore, the Global Fire Assimilation System (GFAS), developed by Max Planck Chemical Institute, Germany (Kaiser et al. 2012), reports the estimations of aerosol, reactive gas, and greenhouse gas emission from wildfires to the atmosphere. GFAS is established on satellite-based fire radiative power (FRP) products from the MODIS instrument present in Terra and AQUA satellites. The emission data are freely available for download at GFAS's website (https:// perma link. aeris-data. fr/ GFASv1.3) and are daily global maps with 0.1° of spatial resolution in the NetCDF file format.
The methodology used by Kaiser et al. (2012) is based on the bottom-up approach. It uses the FRP to estimate the biomass burnt rate, which is used to calculate the emissions using factors for each pollutant (Andreae and Merlet 2001) and distinguishing between eight land use classes. The emissions of particulate matter consider a linear regression with FRP and analyzing the MODIS aerosol optical depth. It is remarkable to distinguish that GFAS has been analyzed and evaluated in several studies (Pereira et al. 2016;Reddington et al. 2016;Nikonovas et al. 2017;Pan et al. 2020).
The GFAS dataset could help to evaluate the regional and global fire emissions on air quality using numerical modeling. Models like CMAQ (Byun and Schere 2006), CAMx (Corporation and Way 2013), and WRF-Chem (Grell et al. 2005) have been used to report the transformation of air pollutants in the atmosphere. These air quality models require an essential input: the emissions data, usually obtained from external software like PREP-CHEM-SRC (Freitas et al. 2011) for WRF-Chem studies and SMOKE (Baek and Seppanen 2018) for CMAQ and CAMx modeling. Those models process the emission inventory and generate the required file format of the specific air quality model.
The usual pollutants studied in the air quality modeling researches include the coarse (PM 10 ) and fine (PM 2.5 ) particulate matter and its submicron-type aerosols like black carbon (BC) and organic carbon (OC). BC is generated in an incomplete combustion process (Petzold et al. 2013), and the accurate effects of this aerosol on climate change can only be estimated with the proper emission inventory (Koch et al. 2009). Also, the health risks and the economic value due to the exposure to PM 2.5 can be performed by air quality modeling studies, and it also depends on the emission inventory involved (Stowell et al. 2019;Lai et al. 2020). Biomass burning is one of the primary sources of BC, OC, and PM 2.5 worldwide (Bond et al. 2013). The adverse effect of BC on climate change and the health impacts of PM 2.5 exposure has led to increased research of these aerosols in the last decades.
Recently, several studies evaluated the effect of biomass burning and wildfires by using air quality models. Table 1 summarizes some publications using diverse wildfire and biomass burning datasets for different regions of analysis. Otherwise, Johnson et al. (2020) reviewed other studies, including dispersion model characteristics. Unfortunately, none of them exposed the processing steps of the emission inventories for air quality modeling, limiting its application on other regions.
Many efforts have been developed during the last decade simulating air pollutants in regional and hemispheric scales to observe long transport in the atmosphere (Huang et al. 2015). The frequency and severity of the wildfires emissions in the most Southern Hemisphere have been associated with particulate matter deposition in the Andes mountains and Antarctica (Shi et al. 2019a;Cereceda-Balic et al. 2020). Unfortunately, the lack of information about this emission source's contribution could be the principal reason for unavailable studies in this region. This study focused on the processing of the biomass burning emissions from GFASv1.3 for future air quality modeling in CMAQ or CAMx. The NetCDF files from this database expose the emissions in the unit kg·m −2 ·s −1 , which is required for processing in SMOKE.   (Zender 2008), and the detailed steps are shown in Table 2. The first step was to extract the information in the period of interest (2014, January 1st to December 31st). SMOKE does not process files with the variable time when the NetCDF file format is used to input the emission inventory.
Step 2 deleted the attribute time in each file generated in step 1. Next, in step 3, the files were changed to classic NetCDF format as a SMOKE requirement. As a result, the formatted files were processed in SMOKE (version 4.5), and the preliminary results are shown in the Supplementary Material ( Figure S1).
The preliminary results obtained in SMOKE showed errors in the input files due to registries in the ocean, which is impossible for biomass burning emission sources. The attribute latitude in the original GFAS1.3 files starts at 89.95° in the north and finishes at −89.95° in the south. However, the required latitude order in SMOKE is opposing, from −89.95° in the north to 89.95° in the south. The solution was made in step 4 when the latitude information was inverted in the file generated in step 3.
Finally, the simulation in SMOKE was made for the year 2014. This study was centered in Antarctica, using polar stereographic projection and considered a hemispheric domain as reported by Pino-Cortés et al. (2020). The processing steps in SMOKE and the postprocessing analysis reported in that publication were also applied to this study. SMOKE reads the daily input files by using the module SMKINVEN as a gridded file for each pollutant of analysis. The next step was the processing in the module SPCMAT associated with the chemical speciation, differencing between particles and gases, according to the user input. Then, the module GRD-MAT was applied to obtain a matrix with the location of the emissions in the domain of analysis. After that, the hourly profiles were obtained for each cell by using the module TEMPORAL. In this study, the emission files have 180 grid cells for each side of the domain and 108 km of horizontal resolution. The hourly profile was set to constant because there is no information about it for this source files were exposed. Finally, the module SMKMERGE used the preliminary files in previous modules and generated the output files required by the air quality models. All steps were replicated for each daily input to obtain 365 files. The command ncrcat, from NCO, was used to merge all outputs in only one file to analyze the emissions.

Results and discussion
Each output file generated in SMOKE4.5 was merged to obtain one file with all the emissions of OC, BC, PM 2.5 , and PM 10 from biomass burning for the year 2014. The highest annual emissions during 2014 of black and organic carbon in the most relevant cities are shown in Table 3.
The individual results of BC and OC exposed in Table 3 cannot be compared to official reports or published papers due to the lack of information. Unfortunately, those species are not included in the traditional emission inventories in the Southern Hemisphere. According to the records in GFASv1.3 datasets and the analyzed outputs obtained in SMOKE, the highest emissions of BC were observed in Maputo, Temuco, and Darwin, in decreasing order. The registries in South America, specifically in big cities like Santiago de Chile, Asunción and Rio de Janeiro, also have high BC emissions. Therefore, the OC registries are incredibly high in Canberra and Maputo, compared to the rest of the cities. The OC generates a cooling effect on the atmosphere (Stocker et al. 2013) and is mainly produced by the condensation of organic vapors from incomplete combustion like biomass burning. That is why the OC/BC and OC/ PM 2.5 ratios are helpful to understand particle light absorption in the atmosphere. The OC/BC ratio in the emissions analyzed is between 6 and 16, and it is in the range reported by Ballesteros-González et al. (2020), but the minimum limit is widely scattered. This ratio could also predict the source of the emission. Lower values (4-5) characterize the open burning emissions of crop residues, the median ratios (6-7) are related to forest fire, and the highest ratios (> 8) are observed in grassland/savanna fire emissions (Qin and Xie 2011).
The values in Table 3 suggest that PM 10 generated in the biomass burning from GFASv1.3 includes 60-75 % of PM 2.5 , and the highest values were in those cities mentioned above. It means a direct relation between particulate emissions and their speciation. The higher PM 2.5 /PM 10 ratio in the biomass burning emissions contributes the source origin to the long range transportation of the particulates and confirms the results obtained in previous modeling studies (Shi et al. 2019a;Cereceda-Balic et al. 2020) about the deposition of the aerosols from biomass burning emissions in zones located several kilometers from its origins, like Antarctica and the Andes Mountain.
According to the annual estimations, Canberra and Maputo registered the uppermost fine and coarse particulate The average black carbon fraction in fine particulate matter emissions resulted mainly in 4%, but this variable could be up to 10%. This result is similar to other studies (Chow et al. 2011) for this source emission. Otherwise, as shown  Table 3 and Figure 1, the OC/PM 2.5 ratio showed a wide range between 47 and 72%. The higher ratio of OC/PM 2.5 produced from the forest and crop residues fires is due to their high fuel load characteristics (Shi et al. 2019b). In contrast, lower values are shown in savanna burning (Bond 2004). This last land type is the majority biomass burned during 2014 according to the OC/BC ratio from GFAS emissions, and it is exposed in Shi et al. (2020) for the same year.
The higher values were located in Eastern Australia, New Zealand, Paraguay, and Northern Argentina. In contrast, lower ratios were observed in Southeast Brazil and Eastern Madagascar. The distribution of those ratios in the domain is shown in Fig. 1.
The temporal profile of the total emissions of pollutants from biomass burning is exposed for eight cities in Fig. 2. This variable is useful for best knowledge of this source emission.
African cities like Maputo and Harare showed their higher emissions from June to September. This profile was also reported by Shi et al. (2020). The emissions from Santiago de Chile, Canberra, and Bahia Blanca occurred up to 95% in January (summer season in the Southern Hemisphere). It could be explained by the direct positive effect of the mean temperature and agriculture on the number of fires and burned areas. Both variables are significant ecological predictors of fire activity (Gómez-González et al. 2019). Those results also could explain the registries in Temuco, where biomass burning emissions were recorded from January to April (included) in 2014 due to the burning of agricultural stubble in the season, typical agricultural practice in rural areas. The study published by Rubio et al. (Rubio et al. 2015) analyzed the nearby wildfires on 4 and 8 January 2014 in Santiago de Chile and exposed the increased concentration of ozone and particulate matter in the urban monitoring stations. The same wildfire was simulated before (Cuchiara et al. 2017) using the WRF-Chem model using the FINN database for the emission inventory. That study reported some limitations using that input associated with the uncertainties.
Finally, there are zones with significant emissions in the entire study domain, as shown in Fig. 3.
The annual emissions were higher in the African continent. Also, the emissions in the North of Australia are highly remarkably. Both regions are affected every year by wildfire, generating higher emissions of many pollutants. In South America, it has highlighted the total annual emissions in the Paraguay and northern region of Argentina, similar to the values reported in Shi et al. (2020). Also, in central-southern areas in Chile, a high amount of emissions are shown. In this country, the highest number of wildfires and burnt hectares occurred in 2014 since 1960 (Úbeda and Sarricolea 2016). Unfortunately, that study did not report the air pollutant emissions, and the effect of this source is not exposed.

Summary
The GFASv1.3 dataset shows widespread biomass burning through the Southern Hemisphere with the largest emissions in South America, the southernmost region of the African continent, and Australia. This study showed a practical method to process the biomass burning files from GFASv1.3 in SMOKE. The NCO commands were applied in four steps to change them to obtain the required NetCDF input format. The results and the methodology exposed could bring important information to Southern Hemisphere countries, especially in highly populated cities with several episodes with high air quality index of pollution. Also, the national emission inventories could be improved, and future air quality modeling could be applied for source effect evaluation.