Airborne measurements of fire Emission Factors for African biomass burning sampled during the MOYA Campaign

. Airborne sampling of methane (CH 4 ), carbon dioxide (CO 2 ), carbon monoxide (CO), and nitrous oxide (N 2 O) mole fractions was conducted during field campaigns targeting fires over Senegal in February and March 2017, and Uganda in 20 January 2019. The majority of fire plumes sampled were close to, or directly over burning vegetation, with the exception of two longer-range flights over the West African Atlantic seaboard, (100 – 300 km from source) where the continental outflow of biomass burning emissions from a wider area of West Africa was sampled. Fire Emission Factors (EFs) and modified combustion efficiencies (MCEs) were estimated from the enhancements in measured mole fractions. For the Senegalese fires, mean EFs and corresponding uncertainties in units of g per kg of dry fuel were 1.8 (± 0.06) for CH 4 , 1633 (± 56.4) for CO 2 and 25 69 (± 1.6) for CO, with a mean MCE of 0.94 (± 0.005). For the Ugandan fires, mean EFs (in units of g kg -1 ) were 3.1 (± 0.1) for CH 4 , 1610 (± 54.9) for CO 2 and 78 (± 1.9) for CO, with a mean modified combustion efficiency of 0.93 (± 0.004). A mean N 2 O EF of 0.08 (± 0.002) g kg -1 is also reported for one flight over Uganda; issues with temperature control of the instrument optical bench prevented N 2 O EFs from being obtained for other flights over Uganda. This study has provided new datasets of African biomass burning EFs and MCEs for two distinct study regions, in which both have been studied little by aircraft 30 measurement previously. These results highlight the important intracontinental variability of biomass burning trace gas and can be used to better constrain future biomass burning emission budgets. More generally, these results highlight the importance of regional and fuel-type variability when attempting to spatially scale biomass burning emissions. Further work to constrain EFs at more local scales and for more specific (and quantifiable) fuel types will serve to improve global estimates of biomass burning emissions of climate-relevant gases. This paper presents the results of airborne surveys conducted over regions of Senegal and Uganda with high prevalence of biomass burning events. Two aircraft-based field campaigns, using the UK Facility for Airborne Atmospheric Measurements 100 Atmospheric Research Aircraft (FAAM ARA), were conducted in widely separated parts of Northern Sub-Saharan Africa as part of the MOYA project. The first was based in Senegal between 27 February 2017 and 3 March 2017, and the second based in Uganda between 16 January 2019 and 30 January 2019 (henceforth referred to as MOYA-I and MOYA-II for the 2017 and 2019 campaigns respectively). 110 N 2 O dry-air mole fractions were measured using an Aerodyne Quantum Cascade Laser Absorption Spectrum (QCLAS) as described by Pitt et al. (2016). This instrument uses a single thermoelectrically cooled quantum cascade laser tuned to a wavelength of ~4.5 µm. The QCLAS is calibrated using three calibration gas standards, all of which are traceable to the World Meteorological Organisation (WMO) X2006 calibration scale for N 2 O. A 1σ uncertainty of 0.58 ppb was estimated for 1 Hz N 2 O mole fraction measurements during the MOYA-II flights. We only report data for the MOYA-II (Uganda) campaign in this study as this instrument was not fitted to the aircraft during the MOYA-I (Senegal) campaign. that the linear interpolation of in-flight calibrations yields a mean bias <1 ppb with a 2 sigma precision of ppb at 150 ppb for 1 Hz CO measurements, optimally. we faulty may our CO 2017-19, yielded ± 9 ppb bias in our data. The potential impact of this positive bias is further detail by Priestley et al ., (2018a; 2018b) for ground-based deployment has recently been modified and certified for use on the FAAM ARA and was used for real time detection of HCN and HNCO in this study. The instrument and its subsequent modification is described in detail here, as this study presents the first measurements from the modified ToF-CIMS aboard the FAAM ARA. The original instrument was manufactured by Aerodyne Research Inc. and employs the ARI/Tofwerk High 220 Resolution Time of Flight Mass Spectrometer. Briefly, iodide ions cluster with sample gasses creating a stable adduct that is analysed using time of flight mass spectrometry, with an average mass resolution of 4000 (m/∆m). The inlet design was based on the configuration characterised by Le Breton et al., (2015), an atmospheric pressure, rearward facing, short residence time inlet, consisting of a 3/8” diameter polytetrafluoroethylene (PTFE) tubing with a total length to the 225 instrument of 48 cm. A constant flow of 12 SLM is mass flow controlled to the ion-molecule reaction region (IMR) using a rotary vane pump (Picolino VTE-3). 1 SLM is then subsampled into the IMR for measurement. An Iris system as described by Lee et al., (2018) was then employed to pressure and mass flow control the sample flow into the instrument, avoiding sensitivity budget, and the work demonstrates the importance of good knowledge of fuel carbon and moisture content for 610 the accurate reporting of EFs. This study demonstrates the utility of airborne measurements for characterising biomass burning emissions from multiple fires over wide areas. Further work is required to investigate the link that fuel carbon and fuel/soil moisture content may have on the emission of methane from biomass burning.

changes that would be associated with large variations in pressures in flight that is not controlled sufficiently by the constant flow inlet. This works upon the principle of the manipulation of the size of the critical orifice in response to changes in the IMR 230 pressure. As with the Lee et al., (2018) design, this works by having a stainless-steel plate with a critical orifice and a movable PTFE plate on top of this, also with a critical orifice. These orifices either align fully and allow maximum flow into the instrument or misalign to reduce flow. This movement is controlled by the 24VDC output of the IMR Pirani pressure gauge in relation to the set point and the control unit was designed collaboratively with Aerodyne Research Inc. The IMR set point was 80 mbar for the MOYA campaign, which is set through a combination of pumping capacity on the region (Agilent IDP3), mass 235 flow-controlled reagent ion flow and sample flow. The reagent ion flow is 1 SLM of ultra-high purity nitrogen mixed with 2 SCCM of a pressured known concentration gas mix of CH3I in nitrogen, passed through the radioactive source, 210 Po. The total flow through the IMR is measured (MKS MFM) at the exhaust of the Agilent IDP3 pump so that not only the IMR pressure is monitored but the sample flow also. All mass flow controllers and mass flow meters are measured and controlled using EyeOn.
The 1σ variability in the IMR pressure during MOYA is 4% and 6% in the sample flow. 240 A standard Aerodyne pressure controller is also employed on the short segmented quadrupole (SSQ) region, with two purposes, easily setting the required pressure during start up but to also make subtle adjustments in this region should the IMR pressure change significantly. This works upon the principle controlling an electrically actuated solenoid valve in a feedback loop with the SSQ pressure gauge to actively control a leak of air into the SSQ pumping line. The SSQ is pumped using Ebara PDV 250 245 pump and held at 1.8 mbar. The 1σ variability in the SSQ pressure during MOYA is <1%.
Instrument backgrounds are programmatically run for 6 seconds every minute for the entire flight, by overflowing the inlet at the point of entry into the IMR. Here a 1/16 th PTFE line enters through the movable PTFE top plate, ensuring that the flow exceeds that of the sample flow. Inlet backgrounds are often run multiple times during flights manually by overflowing as close 250 to the end of the inlet as possible with 20 SLM. Data is taken at 4Hz during a flight, which is routinely averaged to 1 Hz for analysis. Of the 6 points in each background, the first 2 and last point are unused and the mean of the background is calculated using custom python scripting. Using linear interpolation, a time series of the instrument background is determined, humidity corrected if required and then subtracted to give the final time series of each measured mass. Instrument sensitivity to increased humidity changes influences the sensitivity of the instrument to HCN and corrections are applied here to correct both the 255 instrumental backgrounds and final time series of HCN reported here. Only qualitative HCN and HNCO data is reported here as quantitative data is not required for the approach of plume identification used in this study.
The FIGAERO-CIMS instrument analysis software (ARI Tofware version 3.1.0) was utilized to attain high resolution, 1Hz, time series of the compounds presented here. For the UMan CIMS, mass-to-charge calibration was performed for 5 known masses; I-, I-.H2O, I-.HCOOH, I2-, I3-, covering a mass range of 127 to 381 m/z. The mass-to-charge calibration was fitted to a 3rd order polynomial and was accurate to within 2 ppm. HCN and HNCO in this case were identified with a 1 ppm error.
Whole air sample (WAS) were collected onboard the aircraft in 3L silica passivated stainless steel canisters (Thames Restek, UK). Sample collection was triggered manually to sample within and outside of fire plumes, guided by the real time methane 265 measurements from the FGGA onboard and visual identification of when the plumes were being crossed. Fill times when sampling the fire plumes were 20 seconds, representative of an integrated air sample over a 2 km track. Methane mole fraction in the WAS flasks was measured in the Royal Holloway greenhouse gas laboratory using a Picarro 1301 cavity ringdown spectroscopy analyser, and methane isotopic analysis (δ 13 C) was carried out by gas chromatographyisotope ratio mass spectrometry using a Trace Gas preconcentrator and Isoprime mass spectrometer (see Fisher et al., 2006 for details of the 270 technique).

Calculation of Emission Ratios and Emission Factors
In order to select when sampled air was influenced by biomass burning emissions, hydrogen cyanide (HCN) and CO were 275 used as biomass burning tracers. HCN was chosen as it is almost exclusively emitted from biomass burning, representing 70-85% of the total global HCN source (Li et al. 2003) and has a sufficiently long atmospheric lifetime (relative to advection timescales prior to sampling) of 2-4 months, making HCN a suitable inert tracer for characterising biomass burning plumes (Li et al. 2000).
Like HCN, significant amounts of CO, which has an atmospheric lifetime of 1-3 months (Ehhalt et al. 2001), are emitted from 280 biomass burning. CO is also emitted by vehicles, primarily petrol-fuelled and less so by diesel. However, it is likely that biomass burning is the dominant source of carbonaceous emissions in rural areas of Africa as studied here, whereas vehicular carbon emissions are likely concentrated towards urban centres (Gatari et al. 2003). HCN was used as a biomass burning tracer for the MOYA-II (Uganda) analysis. However, as the ToF-CIMS was not fitted to the aircraft during the MOYA-I campaign, no HCN measurement is available for this dataset, and hence CO is used as the biomass burning tracer for MOYA-I analysis. 285 In order to quantify biomass burning emissions from the enhancements in trace gas mole fraction seen in fire plumes, Emission Ratios (ERs) and EFs were calculated for each species in each fire plume. In this case, an ER is defined as the ratio of a species X relative to a reference species Y. The reference species chosen for this work was CO, as it is relatively inert in the timescale of these measurements, had a relatively stable regional background concentration during these campaigns, and in these rural 290 field areas is almost exclusively emitted during combustion processes and not by other sources such as vehicles (Andreae and Merlet. 2001). The expression for ER calculation is shown in Eq. (1).
ERs calculated using this approach are also referred to as Normalised Excess Mixing Ratios (NEMRs). When fresh plumes are sampled close to source, NEMRs can be treated as an ERs, calculated using Eq. (1). However in aged plumes, this approach cannot be used to calculate ER, and NEMR is no longer equal to ER. This is due both to chemical processes within the plume that can change composition as well as mixing of background air into plume air (Andreae and Merlet. 2001;O'Shea et al. 300 2013;Yokelson et al. 2013). Analyses from MOYA-I and MOYA-II flights display no significant plume ageing with the potential exception of flight C007.Therefore, Eq. (1) can be used confidently to calculate ERs confidently for most flights. The calculation of ERs for flight C007 is discussed further in Sect. 4.2.
In order to calculate ERs for near-field biomass burning plumes, a baseline mixing ratio (Xbackground) was calculated as the 305 average mixing ratio over 10 seconds of sampled data to either side of each detected plume. The same baseline data periods chosen for each plume were used for all gas species, to ensure that ERs were comparable and not influenced by inconsistent baseline criteria. The area under the plume was then determined by integrating the peak in the concentration versus time data series, giving a total plume concentration (Xplume). These values were then used in Eq. (1), along with the corresponding values for CO, to determine an ER. Due to the absence of individual sharp enhancements resolved for specific fire plumes in the far-310 field flights, a least-squares linear regression of all in-plume points of X versus in-plume points of CO is used to determine ERs for the far-field flights. The ER is equal to the slope of this linear regression.
Using the calculated ER for each species, EFs were calculated using the carbon mass balance technique (Ward et al. 1984;Radke et al. 1988) An EF is defined as the mass of species emitted (in grams) per kilogram of dry matter burnt. The expression 315 for calculating emission factor is given in Eq. (2).
where is the mass fraction of carbon in the dry fuel. This is typically between 0.45 and 0.55 for biomass. A value of 0.475 was assumed in this work to best represent African biomass burning (Cofer et al. 1996;Ward et al. 1996). is the molecular weight of species X and is the atomic mass of carbon-12. The term is the molar ratio of species X to total carbon in the plume, which is calculated using Eq. (3).
In Eq. 3, total carbon in the fire plume was assumed to be the sum of CO, CO2 and CH4 emitted. However, as all carboncontaining species could not be measured in this study, the total carbon present in the plume may be underestimated by 1-2% (as reported by Yokelson et al. 1999).
A statistical threshold approach was used to determine when a biomass burning plume was sampled during flights. For flights 335 where HCN measurements are available, HCN enhancements exceeding seven standard deviations above the local background were used to select data for ER and EF calculation. Where HCN was not available during MOYA-I, a CO threshold of seven standard deviations over the local background concentration was used. For the far-field flights during MOYA-I (C006 and C007) CO mixing ratios exceeding 15 standard deviations above the local background were chosen for analysis.

Modified Combustion Efficiency
In addition to EF, the modified combustion efficiency (MCE) is another useful parameter that can be calculated for each biomass burning plume. MCE is here defined by Eq. (4).

345
350 MCE can be used to determine the degree to which a fire is smouldering or flaming (Ward and Radke, 1993 It is therefore useful to investigate the trend between EF and MCE for different fire plumes (Urbanski, 2013). In the following 355 section, we calculate EFs and MCEs for sampled fire plumes in the MOYA-I and MOYA-II campaigns.

360
The standard error of the mean (SE) and the mean measurement uncertainty (MU) are reported for each mean EF and MCE displayed in Table 1, The SE here is determined from all EF and MCE calculated for a single flight, and represents the variability of EF and MCE within a flight. The MU is propagated from the instrument uncertainties, therefore each EF and MCE from each fire plume sampled has a measurement uncertainty associated with it. The MUs reported in Table 1 are the average of all individual MUs for all fire plumes sampled during a given flight. 365 ERx is calculated using Eq. (1) by subtracting CObackground from COplume; any CO measurements systematic positive offset would therefore cancel out and not affect the uncertainty of ERx. The detection of COplume during MOYA-I is based on the exceedance of either seven or fifteen standard deviations above background; a CO measurements offset on the background may therefore affect this data filtering step; however due the wide dynamic range of CO measurements encountered during the plumes 370 sampling, we believe a bias will have a very minimal effect on the filtered plume data set used in our analysis. Similarly, the calculations of EFx using Eq. (2) and (3), and MCE using Eq. (4), rely on ΔCO, which is unaffected by CO measurement bias as previously stated.  and to a lesser extent, C004, show a linear relationship between points with lower inverse CH4 mixing ratio (enhanced CH4) and points with more enriched δ 13 C-CH4 signatures. This suggests that biomass burning emissions were captured by whole-air sampling during these flights. The intercept of this linear regression provides information about the specific fuel types of the fire plumes sampled. The intercept of -33.7 ± 1.1 ‰ suggests the fuel is C3 forest litter (Brownlow et al., 2017, Dlugokencky et al. 2011, Chanton et al 2000, as opposed to C4 tropical grasses and maize or millet and sorghum crop waste. Unfortunately, 415 flights over mixed sources in Uganda meant that Keeling plot analysis could not be used to determine the isotopic composition of fire emissions in the same way as carried out for Senegal.  However the methane EFs for C004 and C005 (2.3 ± 0.08 g kg -1 and 1.4 ± 0.05 g kg -1 respectively) in this region, at the northern fringe of the African moist tropics, is more comparable to the savannah and grassland methane EF (2.7 ± 2.2 g kg -1 ) averaged from multiple previous studies by Andreae. (2019). Additionally, Mean CO EFs (84 ± 2.0 g kg -1 for C004 and 61 ± 1.5 g kg -1 for C005) are also more comparable to the savannah and grassland CO EF of 69 ± 20 g kg -1 than the tropical forest analysis focus on burning associated with Amazonian deforestation, which consists mostly of broad-leafed evergreen forest.
In contrast, the Casamance region consists of facultatively deciduous broad-leafed forested savannah, which was observed from the aircraft. It is thus possible that any forest matter burned during the MOYA-I flights consists of dry leaf-litter fuel, whereas the Andreae (2019) study comprising mostly Amazonian land clearing may have included burning of whole evergreen tree structures. In addition to this, the modified combustion efficiencies of the C004 and C005 fires (0.93 ± 0.0031 and 0.95 ± 435 0.0030 respectively) are both higher than that reported in Andreae (2019) for tropical forest (0.91 ± 0.03), and are more comparable with the Andreae. (2019) MCE for savannah and grassland burning (0.94 ± 0.02). This is likely due to the lower fuel moisture content of dry leaf-litter and savannah grasses, as opposed to Amazonian evergreen. Therefore the difference in MCE and the difference in the specific fuel mixture are thus the likely causes for this discrepancy between EFs observed here and those in Amazonia as reported by Andreae (2019). 440

Flight C132
Flight C132 was undertaken on 28 January 2019, as a survey of the Lake Kyoga wetland area. Two crop waste biomass burning plumes were sampled from two distinct fires in the area (see Fig. 2). A time series of various trace gas mixing ratios during 445 this flight is shown in Fig. 5.
As seen in Fig. 5, enhancements (relative to background) in all trace gases were observed in the two biomass burning plumes.
However, N2O mixing ratio data during the two sharp enhancements were discarded due to aircraft turbulence, which may have corrupted data quality. As a result of the discarded data, as well as instrument drift owing to malfunction of the laser 450 coolant system, N2O EFs are not reported for flight C132. Fig. 6 shows the land cover of Uganda where the fire sampling flights were carried out. In agreement with on-board observations from the aircraft, much of the land surrounding Lake Kyoga is classified as cropland, and the fuel for the fires appeared to be primarily crop waste. This is a major farming region, with the main crops including maize (a C4 plant) and 455 cassava (C3) south of Lake Kyoga, and sorghum (C4) north of the Lake. (FEWS NET. 2019). The mean EFs calculated for C132 (5.2 (± 0.18) g kg -1 for CH4, 1554 (± 52.9) g kg -1 for CO2 and 109 ± 2.6 g kg -1 for CO) agree within overlapping uncertainty with mean agricultural burning EFs of 5.7 ± 6.0 g kg -1 for CH4, 1430 ± 240 g kg -1 for CO2, and 76 ± 55 g kg -1 for Consequently, Fig. 7b shows methane emission factor decreasing with lower ΔHCN/ΔHNCO ratio. This further affirms that difference in combustion completeness is the primary driver of methane EF variability observed during MOYA-II.
Unfortunately, A similar analysis could not be carried out for MOYA-I as the ToF-CIMS was not fitted to the aircraft during the MOYA-I flights. 495 As in flight C132, N2O measurements for flight C133 were unreliable and data were discarded due to the effects of aircraft motion on the instrument optical bench during turbulence. Furthermore, issues with the temperature control of the QCLAS optical bench meant that the baseline noise and drift of the N2O signal increased during this flight. This resulted in a reduced signal-to-noise ratio of N2O in the plume. For these reasons, an N2O EF is not reported for flight C133. However, optical bench 500 temperature control was adequate during flight C134, and aircraft turbulence did not impact N2O data quality significantly during sampling of some fire plumes. Hence calculation of N2O EFs was possible for six of the nine fire plumes sampled during flight C134.
In general, the N2O mixing ratio enhancements in the fire plumes are small (<10 ppb) relative to the background variability (and instrumental noise) of the N2O dataset (up to 2 ppb). Hence the signal-to-noise ratios of the in-plume N2O enhancements 505 are poorer than the in-plume enhancements of other species. As a result of this, the uncertainty relative to the mean N2O EF for C134 is larger than those seen in the other species measured. Despite the combination of instrument issues and poor signalto-noise ratio, the N2O EF for flight C134 (0.08 (± 0.003) g kg -1 ) agrees within overlapping uncertainty with the savannah fire N2O EF reported by Andreae (2019) (0.17 (± 0.09) g kg -1 ) 510 Fig. 8 shows a strong linear relationship between MCE and CH4 EFs for both MOYA-I and MOYA-II, with R 2 values of 0.867 and 0.991, respectively. It is worth noting that CH4 EFs and corresponding MCE for the far-field flights C006 and C007 are not included in Fig. 8, as the EFs from these flights are representative of multiple fires with a mixture of phases, whereas the near-field EFs are representative of single fires with a single combustion efficiency associated with them. This trend is expected as higher MCE, and hence more complete flaming combustion, would lead to increased emission of more oxidised combustion 515 products (CO2) and less emission of more reduced compounds such as CH4. Despite this, there is a significant contrast in slope and intercept between the MOYA-II and MOYA-I linear regressions. The difference in the linear regressions could possibly be accounted for by probable differences in carbon content between the Senegalese and Ugandan fuel mixtures. However, due to the lack of carbon content determination for the biomass burned in this study, and with the likelihood of the fuel source being mixed, the effect of differing carbon content is difficult to quantify. An additional hypothesis is that higher average soil 520 moisture in northern Uganda compared to south-west Senegal could result in soil parching and consequent release of methanerich air from the soil surrounding wildfires, however more work is required to investigate if soil moisture could affect wildfire methane EFs in this way. time as shown in Fig 1. 550 Fig. 12 shows that the biomass burning emissions sampled during C007 originated from different regions of West Africa, and have a wider age range. With these complex air masses, the approximate age of the biomass burning emissions observed in C007 was estimated to be older than that in C006, with an approximate age of 1-2 days. Consequently, the emissions sampled in C007 were representative of a wider area of West African biomass burning, spanning from south-west Senegal down to 555 Sierra Leone. Due to the significantly older plume age of the C007 biomass burning emissions, it is possible that significant chemical aging and/or mixing of background air with plume air has occurred, and hence the ERs or EFs derived from this

Conclusion
Airborne observations of CH4, CO2, and CO emissions from biomass burning were carried out in southern Senegal in February/March 2017 and northern Uganda in January 2019. Mean EFs of 1.8 (± 0.05) g kg -1 for CH4, 1633 (± 56.4) g kg -1 for CO2 and 69 (± 1.7) g kg -1 for CO were obtained from the Senegalese fires, with a mean modified combustion efficiency of 0.94 (± 0.005). Mean EFs of 3.1 (± 0.1) g kg -1 for CH4, 1610 (± 54.9) g kg -1 for CO2 and 78 (± 1.8) g kg -1 for CO were obtained 595 for the Ugandan fires, with a mean modified combustion efficiency of 0.93 (± 0.004). A mean N2O EF of 0.08 (± 0.002) g kg -1 is also reported for six fire plumes sampled over Uganda. CH4 EFs showed strong linear relationships with modified combustion efficiency for both Senegal and Uganda, with R 2 values of 0.867 and 0.991, respectively. The variability in EFs within each study area was attributed to the mixed-phase nature of the fires, with a range of combustion efficiencies observed.
These results also suggest that Ugandan fires have a higher methane emission factor for the equivalent combustion efficiency 600 observed for Senegal. This may be a consequence of the difference in fuel carbon content between the Ugandan savannah grass and cropland waste fuels, and the Senegalese forest litter and grassland fuel. This highlights the importance of considering both regional and local variability when attempting to spatially scale biomass burning emissions, and suggests that singular regional EF values may lead to inaccurate estimates. Further work to constrain EFs at more local scales and for more specific (and quantifiable) fuel types will serve to improve global estimates of biomass burning emissions of climate-relevant gases. 605 This work demonstrates the value of airborne measurements for characterising biomass burning emissions from multiple fires over wide areas. This study has provided unique in situ datasets in two geographical regions where there has hitherto been little study by aircraft measurement. The results will improve understanding of the role of African biomass burning in the global carbon budget, and the work demonstrates the importance of good knowledge of fuel carbon and moisture content for 610 the accurate reporting of EFs. This study demonstrates the utility of airborne measurements for characterising biomass burning emissions from multiple fires over wide areas. Further work is required to investigate the link that fuel carbon and fuel/soil moisture content may have on the emission of methane from biomass burning.      https://doi.org/10.5194/acp-2020-558 Preprint. Discussion started: 10 July 2020 c Author(s) 2020. CC BY 4.0 License.