Particulate emissions from biomass burning can both alter the atmosphere's radiative balance and cause significant harm to human health. However, due to the large effect on emissions caused by even small alterations to the way in which a fuel burns, it is difficult to study particulate production of biomass combustion mechanistically and in a repeatable manner. In order to address this gap, in this study, small wood samples sourced from Côte D'Ivoire in West Africa were burned in a highly controlled laboratory environment. The shape and mass of samples, available airflow and surrounding thermal environment were carefully regulated. Organic aerosol and refractory black carbon emissions were measured in real time using an Aerosol Mass Spectrometer and a Single Particle Soot Photometer, respectively. This methodology produced remarkably repeatable results, allowing aerosol emissions to be mapped directly onto different phases of combustion. Emissions from pyrolysis were visible as a distinct phase before flaming was established. After flaming combustion was initiated, a black-carbon-dominant flame was observed during which very little organic aerosol was produced, followed by a period that was dominated by organic-carbon-producing smouldering combustion, despite the presence of residual flaming. During pyrolysis and smouldering, the two phases producing organic aerosol, distinct mass spectral signatures that correspond to previously reported variations in biofuel emissions measured in the atmosphere are found. Organic aerosol emission factors averaged over an entire combustion event were found to be representative of the time spent in the pyrolysis and smouldering phases, rather than reflecting a coupling between emissions and the mass loss of the sample. Further exploration of aerosol yields from similarly carefully controlled fires and a careful comparison with data from macroscopic fires and real-world emissions will help to deliver greater constraints on the variability of particulate emissions in atmospheric systems.
Atmospheric aerosol particles emitted from biomass burning have a substantial influence on global climate, atmospheric chemistry and cloud processes, in addition to being detrimental to human health. Domestic fires used for cooking and heating, agricultural burning, forest wildfires and savannah fires all contribute to the atmospheric loading of biomass burning aerosol (BBA). Andreae and Merlet (2001) identify biomass burning to be the largest source of primary fine carbonaceous particles in the atmosphere globally. Eighty per cent of global BBA is thought to originate from the tropics (Hobbs et al., 1997) and a significant proportion of this is anthropogenic. However, with the resurgence of biomass burning as a source of renewable energy in the developed world (Johansson et al., 2004), issues surrounding BBA are of increasing concern worldwide.
Black carbon and organic carbon are both produced in high quantities by biomass burning, along with inorganic species such as sulfates, nitrates and potassium. Black carbon absorbs incoming short-wave radiation, re-emitting it at infrared frequencies that warm the surrounding air. In contrast, the majority of organic carbon and inorganic species scatter incoming sunlight. Given the presence of both absorbing and scattering aerosol, it is clear that BBA must be well characterised in order for its impact on climate to be fully understood. The ratio of organic matter to black carbon is likely to have a large influence on the net effect of BBA in the atmosphere.
One of the greatest anthropological sources of black carbon is the residential burning of solid fuels, which contributes 25 % of emissions: more than both diesel engines and industrial coal burning (Bond et al., 2013). Given this large contribution, it is important to be able to constrain the estimate of emissions from this source in order to model black carbon in the Earth's atmosphere accurately. A considerable number of studies have been carried out in order to establish the emission factors – the mass emitted per unit fuel burned – of various chemical species for different fuels and circumstances. There has been a particular focus in the literature on emission factors from cook stoves (e.g. MacCarty et al., 2008; Roden et al., 2009; Zhang et al., 2000) and wildfires (e.g. Christian et al., 2007; Yokelson et al., 2008). Others, including Schneider et al. (2006), Weimer et al. (2008) and Elsasser et al. (2013) have investigated the chemical composition of BBA emissions in real time during biomass combustion in the laboratory. Attempts to compile all available measurements of emission factors for a number of different species, both gaseous and particulate, have been carried out by Andreae and Merlet (2001) and more recently by Akagi et al. (2011).
However, due to the complex nature of biomass combustion, and high sensitivity to the combustion environment (i.e. ventilation and heat transfer), it has proven difficult to establish repeatability in the results of laboratory-based biomass burning experiments. Even small alterations in the burning environment can lead to huge variations in particulate emission factors (Akagi et al., 2011) and as such, the majority of studies have found considerable differences in the observed aerosol emission factors and efficiency of combustion between nominally identical burns. Furthermore, the standard approach in many such studies has been to consider only average emission factors from an entire combustion event, which can mask significant changes in emissions that take place throughout combustion. Nonetheless, some recent studies have been carried out, which consider real-time particulate emissions under more controlled conditions, including Alfarra et al. (2007) and Elsasser et al. (2013).
In ambient studies that measure the mass spectra of organic aerosols,
techniques such as positive matrix factorisation (PMF) are regularly
used to identify and quantify distinct mass spectra with different
properties (Zhang et al., 2011). The multilinear engine (ME-2) uses
pre-identified mass spectra to constrain the factors identified by PMF
(Paatero, 1999; Canonaco et al., 2013). Biomass burning is known to
produce one of the most difficult organic signatures to constrain
using this method. This is due to its large variability. For example,
Crippa et al. (2014) advise allowing for a variability of
In this study, an emphasis was placed on enhancing the repeatability of combustion events in order to study aerosol emissions on a mechanistic level, rather than attempting to simulate real-world combustion. In order to achieve this, wood samples were prepared to be as similar as possible during each test. The Fire Propagation Apparatus (ASTM, 2000) was used to control the combustion environment. Samples were heated by a series of infrared heaters and controlled air flow was delivered from below the sample.
Measurements of the combustion products were made in real time using
an Aerosol Mass Spectrometer (AMS) (Aerodyne Research, Inc.,
Billerica, MA, USA) and a Single Particle Soot Photometer (SP2)
(Droplet Measurement Technologies, Longmont, CO, USA) to measure
organic aerosol (OA) and refractory black carbon (rBC)
respectively. Emission factors were measured with a time resolution of
5
Large differences were observed in the ratio between OA and rBC during different periods of each individual burning event. As such, three different phases of combustion were defined based on this ratio, each of which displayed a unique chemical signature.
Combustion experiments were carried out in the Rushbrook Fire
Laboratory, School of Engineering, University of Edinburgh. The Fire
Propagation Apparatus (FPA) allows determination and quantification of
material flammability characteristics including time to ignition,
gaseous emissions, burning rate and heat release rate. Small
(approximately 100
A custom-made stainless steel basket was used to hold the samples and
allowed an air flow to enter from beneath. A flow controller allowed
control over the quantity of air reaching the sample from below; here,
this was held at either 50 or 200
Exhaust gases from combustion were collected in a hood before entering
into the exhaust duct, where turbulence encouraged mixing. Air samples
for aerosol measurement were extracted from the exhaust tube from
a forward-facing subsampling inlet. After passing through a filtering
condensing system, concentration of oxygen, carbon dioxide and carbon
monoxide in the exhaust gas were measured using non-dispersive infrared techniques (Servopro 4200). A separate sample system allowed
aerosol properties in the exhaust to be monitored in real time by the
AMS and the SP2, after being passed through a system of two Dekati
DI-1000 ejector diluters, which diluted the aerosol in pure nitrogen
by a combined factor of 100. These were included to prevent saturation
of the signals. The
The use of two
The diluters used have been designed to sample combustion particles and therefore to minimise particle losses. However, it must be recognised that there may be some perturbation of aerosol composition through processes such as the repartitioning of OA species or diffusional loss of smaller particles. The experiment design has minimised these effects as much as possible.
An Aerodyne Compact Time-of-Flight Aerosol Mass Spectrometer (AMS) was used to measure the OA concentration, as well as that of inorganic species including sulfates, nitrates and ammonium. This instrument has been documented extensively in the literature (see e.g. Drewnick et al., 2005; Canagaratna et al., 2007).
During this test series, the AMS was used in the fast mode (Kimmel
et al., 2011), which allowed results to be collected with a much
higher time resolution than the default “MS” mode. The chopper
alternated between 30 s in its open position, allowing particles to
pass through, and 10 s in its closed position to establish the
background. This mode allowed a time resolution of 0.5
The experimental set-up, showing the sample in the FPA and the location of the sample pipes in the exhaust.
An example of one of the wood samples used.
As described by Allan et al. (2004), a fragmentation table was used to
calculate the contribution of distinct chemical species to each
The SP2 was used to measure the concentration of refractory black
carbon (rBC) in the emitted particles. rBC is carbonaceous, insoluble
and vaporises only at temperatures near 4000
The SP2 has often been used to examine the thickness of organic or other non-incandescent materials that form coatings on rBC particles (e.g. Liu et al., 2014, 2017). The presence of a coating enhances the scattering cross section of the particle, which allows the relative thickness of the coating to be determined using the methodology detailed in Taylor et al. (2015) and Liu et al. (2014). This approach was attempted here, but in this case, rBC was often not emitted concurrently with organic material, which is likely to have resulted in extremely low coating thicknesses during these times. As the technique relies on the interpretation of only the leading edge to reconstruct the scattering signal (i.e. as the particle enters the laser but before the coating begins to evaporate), very thin coatings on small particles are difficult to retrieve with this method, as the algorithm only utilises a very small number of data points with a very low signal-to-noise ratio to fit the “tail” of a Gaussian distribution. Here, even the highest retrieval percentages were lower than 30 %, which means that the successful fits were likely biased towards particles with larger coatings. As such, these data cannot be used.
Outline of the burning environments used for the eight different tests carried out.
The wood used in this experiment was rubberwood (
The emission factors presented in Sect. 3.3 refer to the ratio of the
mass of the emitted species to the mass of dry fuel consumed, in
Eight flammability experiments were undertaken, with different heat fluxes and flow rates. These, and the notation used to refer to them, are presented in Table 1.
Due to the emphasis placed on reproducibility during these tests, some aspects differ from real-world biomass burning, and these must be taken into account when comparing these results with emissions reported elsewhere. The heating of samples throughout the experiments allowed greater reproducibility by reducing the effects of the external conditions, and as a result these data are more representative of burning of a larger accumulation of material. Emissions here are sampled in a duct rather than from an ambient environment and as such, there is no effect of mixing with ambient air.
Each test began by initiating the air flow at the desired rate and
instantaneously exposing the samples to the desired heat flux. A small
flame was fixed above the sample, to drive piloted ignition of the
pyrolysis gases when their concentration was sufficient. The pilot
flame was a pre-mixed flame made from an ethylene/air mixture, which
formed a blue conical flame of around 1
Figure 3 shows the time series of aerosol and gas species being
released during a test with 30
Three panels representing results as a function of time from
a single combustion event under low heat (30
Throughout this experiment, noticeable changes were observed in the MCE,
represented in panel (a), which began to decrease gradually from unity after
ignition. Combustion was highly efficient at the beginning of the experiment,
as evidenced by the MCE remaining above 0.98 for the first 400
Figure 3c shows how measurements of rBC and OA varied over the course of the
combustion event. Aerosol measurements began as soon as the sample was
exposed to the heat flux. A clear peak in OA could be seen prior to ignition
as the sample was exposed to heat, which was also observed in the scattering
volume measurement (A in Fig. 3c). This was likely due to the pyrolysis of
the wood. During this process, gaseous pyrolysis products evolve which are
volatile at high temperatures but are likely to condense into aerosol in the
cooler environment of the exhaust. Almost no rBC particles were seen during
this period as a luminous flame is required for these to be produced. The
aerosol mass derived from the volume-convolved number distribution from SP2
scattering intensity signals is on the same scale as the AMS organic mass in
Fig. 3c if a particle density of 1.1
Immediately after the flame ignition, there was an abrupt decrease in the
production of OA, while rBC simultaneously increased, peaking at just over
40
Approximately 450
After 855
Due to the high level of control and the small sample sizes used in this experiment, it has been possible to distinguish very clearly between these different periods of combustion. It is important to note that during real-world burning or larger scale laboratory-based experiments, combustion will be more heterogeneous, with different phases of combustion highlighted here taking place simultaneously and the emissions free to form complex mixtures through mixing, condensation and coagulation.
Given the distinctive contrast in aerosol emissions at different
times, as well as the recurrence of similar features observed during
all tests, the ratio between the OA mass measured by the AMS and the
sum of OA and rBC measured by the SP2,
As is illustrated in Fig. 3d, phase 1 represents pyrolysis immediately
prior to ignition, where OA makes up more than 85 % of the total
particulate carbon emitted. Phase 2 is the period after ignition
during which OA was less than 15 % of particulate carbon produced
(or rBC more than 85 %). Phase 3 is smouldering-dominant
combustion: the period during which organic mass contributed over
60 % to total particulate carbon. As there is generally a swift
transition from one regime to the next, the majority of the time
series during combustion falls within one of these descriptions,
although some tests, particularly those under
Results found in the literature generally support the assertion that
OA dominates during pyrolysis and smouldering and more rBC is emitted
during flaming than smouldering (e.g. Akagi et al., 2011; Kuhlbusch
and Crutzen, 1995; Simoneit et al., 1999). However, in the majority of
experiments and ambient measurements, the proportion of OA emitted
during flaming is significantly higher than the 15 % used to
identify the phase here. In a review of the literature, Reid
et al. (2005) report a summary of carbon apportionment from studies
using thermal emission techniques, with OA and black carbon reported
as mass fractions. Most of the samples were taken from real-world
forest or Savanna fires. Reid et al. (2005) reported the ratio of
BC : OA from 17 flaming studies (MCE
Figure 4 shows the evolution of rBC and OA mass and SP2-derived scattering particle volume throughout each of the eight combustion tests carried out. For each of these experiments, larger figures showing more detail, as in Fig. 3, are included in the Supplement. The recurrence of the same phases described above can be seen in these tests, with the coloured bar beneath each panel highlighting the phase of combustion taking place. The difference in emissions from phase to phase is clearly much larger than the difference between overall emissions released from different tests.
Time series of the particulate properties during each of the
eight tests carried out. The coloured traces show particulate
concentration, as in Fig. 3b, and the dark grey lines
represent the MCE. The MCE scale on the first plot applies to all
plots. The dark grey boxes show the time between the extinction of
the first piece of wood and that of the last. Where the first line
is dotted, the first extinction was not recorded and the time is
therefore an estimation based on the patterns seen across the tests
in MCE and particulate concentrations around the time of extinction.
Where two scored lines are shown at the concentration peak for
a species, this means that the highest concentration for that
species is above the scale for these plots; more information can be
found in the Supplement. The bars beneath each
Nevertheless, differences between the different burning environments can be identified. For example, the peak in OA particle mass due to pyrolysis prior to ignition was noticeably smaller under low heat conditions than high. The influx of a large amount of energy from the infrared heaters under high heat conditions resulted in a faster and more intense period of pyrolysis before the flame ignited, increasing the vaporisation rate of semi-volatile species from the sample.
The shape of the rBC curve differed slightly between environments,
with a far more gradual decrease under
In both
Figure 5 shows the average mass spectra of organic particles from
phases 1 (a) and 3 (b) as a percentage of the total OA released, along with
the difference between the two phases (c). The mass spectrum from phase 2 is
not considered here due to the very low concentrations of OA emitted. The two
spectra shown are dominated by hydrocarbon ion fragments, including
The mass spectra of OA presented as a percentage of total OA
produced during phases 1
The peaks that were more prominent in phase 1 than phase 3 included
the series
An ambient investigation into the urban burning of solid fuels carried out in London in winter 2012 (Young et al., 2015) identified two different mass spectral factors contributing to the solid fuel organic aerosol (SFOA). These were found using positive matrix factorisation (PMF), a technique used to identify different components of organic aerosol contributing to the overall mix in ambient air (Zhang et al., 2011). The data were sampled from a large air mass with a number of contributing sources, of which two were identified to be SFOA. Various explanations for the origin of these two factors of SFOA were considered, with the authors concluding that different burning conditions was a likely reason that two different factors were produced. The difference between these two factors shows a remarkable similarity to the difference between the two combustion phases examined here; results from Young et al. (2015) are reproduced in Fig. 5d alongside the difference spectrum from the two organic phases observed in our controlled experiments.
Given the striking similarities between these two difference spectra,
it is probable that the SFOA factors identified by Young et al. (2015)
are related to the two OA-producing phases of combustion identified
here, pyrolysis and smouldering. Young et al. (2015) noted that the
peaks observed most strongly for the factor SFOA2, which are similar
here to phase 1 (above the
A closer examination of the results by Young et al. (2015) generally supports these associations. There was a significant time correlation between the two SFOA factors, which is consistent with the proposition that they were produced by the same sources during different phases of combustion. In some cases, SFOA2 (associated with pyrolysis) was seen to peak slightly earlier than SFOA1 (smouldering), which could be related to an increase in pyrolysis early during combustion. There was a slight dependency on wind speed and direction, with SFOA1 more likely to originate from the south and SFOA2 from the east or west. This cannot be immediately reconciled with the idea that factors are related to combustion phases, but could be related to different types of burning taking place in different areas.
Comparing the relative contributions of significant
In Fig. 6a, the data collected from phases 1 and 3 from all eight experiments
are projected onto this space, alongside a diagrammatic representation of fresh BBOA
results from Heringa et al. (2012) and averages from two SFOA factors from Young et al. (2015). The data collected
during this project are slightly closer to the ambient triangle than that of
Heringa et al. (2012), but show a broadly similar pattern, with phase 1 data
containing less
A second comparison of
The two phases of the burn are separated horizontally in
Fig. 6b. Phase 1 occupies the lowest, left-most part of the triangle,
with low values of
The ratio of levoglucosan to potassium has been used in the past as an indicator of biomass combustion efficiency (Harrison et al., 2012). In high temperature fires, levoglucosan is more completely consumed by the flame and more potassium is emitted in particles, resulting in a low levoglucosan : potassium ratio from more efficient combustion. Young et al. (2015) found a lower ratio in SFOA1 (smouldering) than SFOA2 (pyrolysis), leading to the conclusion that the SFOA1 factor was from more efficient combustion.
However, Young et al. (2015) did not consider the possibility that
a factor could be associated with pyrolysis rather than flaming or
smouldering combustion. In our results, the lower levoglucosan during
pyrolysis (judged from
The clear similarities between the two SFOA factors from Young et al. (2015) and the mass spectra of pyrolysis and smouldering combustion presented here give a strong indication that our results could have an application in identifying ambient BBOA factors. Results presented here not only confirm that different phases contribute to BBOA variability by producing different mass spectra; further, they imply that it is possible to distinguish between these two factors in ambient measurements. If future work is carried out to characterise the mass spectra produced by pyrolysis and smouldering combustion more completely, it may be possible to decrease the uncertainty in the BBOA factor constraints by assuming two related but independent biomass burning factors, rather than the current approach of using one.
Ambient studies have been carried out by both Zhou et al. (2017) in the western United States and Brito et al. (2014) in South America, which used PMF to establish three and two BBOA factors respectively. However, analysis of these factors in both cases suggested that the differences between them were due to the aerosol ageing in the atmosphere, rather than any difference in the phase of combustion at the source. Given that there was no ageing of the aerosol in our experiment, the features of phases 1 and 3 described here would likely both have contributed to the “fresh BBOA” factor in each of these ambient studies. The relative remoteness of the sources and warmer temperatures for the measurements from Zhou et al. (2017) and Brito et al. (2014) when compared with those of Young et al. (2015) likely contributed to the inability to distinguish between pyrolysis and smouldering emissions in these cases. This suggests that results presented here are likely to be most valuable in situations where measurements are made in close proximity to the biomass burning source and where atmospheric processing of aerosol is limited. The study of Zhou et al. (2017) was of a summertime wildfire, and that of Brito et al. (2014) was of open biomass burning, in contrast to the observations by Young et al. (2015), which were characterised by household heating fires. Thus, it is possible that the nature of the combustion in these different circumstances played a role in the emissions released.
Previous experiments have used the ratio of
In atmospheric models, the emissions from wildfires, cook stoves and other sources of BBA are often represented by average emission factors (EFs), which relate the production of a given pollutant to the mass of fuel that has been burned (e.g. Liousse et al., 2010). The average EFs employed by these models are measured in the field or in laboratories for specific fuels and, occasionally, burning conditions. They are compiled into inventories, such as those presented by Andreae and Merlet (2001) and Akagi et al. (2011). However, very large variation in EFs can be found even within individual datasets; uncertainties are attributed to fuel type and characteristics, combustion phases, fire intensity and the naturally chaotic behaviour of fires (Reid et al., 2005). Results presented here provide some insight into some of the mechanisms that result in this high variability.
In order to compare results collected here with those in the literature, average EFs for the duration of each phase of combustion have been calculated for OA and rBC for each of the experiments. These have been calculated based on aerosol emissions and the measured mass loss during each phase, as outlined in Sect. 2, with a correction of 100 made to account for dilution in the aerosol sample line. These, alongside average MCEs for the entire combustion period, are displayed in Table 2. The average OA and rBC emission factors for each entire burn are plotted against overall MCEs in Fig. 7.
Emission factors for OA and rBC during each phase for each test. EFs are shown based on the measured mass loss during each phase, and a weighted average is shown in Fig. 7. Errors represent the standard error of the means.
Overall OA and rBC emission factors for each test.
Within each experiment type and phase shown in Table 2, the EFs for
each species were reasonably consistent; the variation from the mean
in 13 cases out of 16 was less than 25 %. It is important to note,
however, that these sample sizes were extremely small (
Considering the entire combustion period instead of individual phases,
Fig. 7a shows a negative correlation between the overall OA emission
factor and MCE, with both being strongly influenced by the burning
environment in which combustion took place. Low heat and high flow
(
The trend in emission factors for rBC is less clear. Previous
literature suggests that a positive correlation can be expected
between rBC emissions and MCE (see e.g. Andreae and Merlet, 2001;
Christian et al., 2003; McMeeking et al., 2009), due to the
preferential emission of rBC during more efficient flaming
combustion. Such a correlation could be seen here for tests at a high
flow rate (
Andreae and Merlet (2001) reported average EFs for biofuel burning
under ambient conditions of 4.0 and 0.59
Due to the high time resolution of the experiments carried out here,
it was possible to examine the change in EFs in more detail throughout
each phase. The change in rBC and OA concentrations with the rate of
mass loss for experiment
OA
There was little correlation between emissions of OA and the mass loss rate of the sample during either phase 1 or 3, the two phases when the vast majority of OA was emitted. This illustrates the highly variable nature of smouldering combustion, even within these well-controlled experiments. Very low rates of mass loss during pyrolysis yielded extremely high and variable OA emissions, which suggests that mass loss rate during this phase is a particularly poor indicator of expected emissions. These results show that the most reliable indicator of the OA emitted from combustion is the efficiency of combustion and the duration of phase 3.
The rBC emissions always increased with an increasing rate of mass
loss. An interesting feature is that above a threshold rate of mass
loss in each experiment, the rate of rBC emission substantially
increased. In Fig. 8 this transition can be seen at a rate of mass
loss of approximately 0.15
The presentation of three phases here, rather than the standard two, provides some insight into the large variation in EFs reported in the literature and the mechanisms producing these. EFs are typically examined in relation to two variables: the fuel mass loss and the MCE, which is used as a proxy for the proportion of flaming or smouldering combustion. Nevertheless, studies have shown that less than 50 % of the variance in particle EFs can be accounted for by the MCE (Ferek et al., 1998; Janhäll et al., 2010; Reid et al., 2005). Comparable uncertainties are seen in relating emissions directly to mass loss (e.g. Freeborn et al., 2008; McMeeking et al., 2009).
The pyrolysis phase as defined here cannot be parameterised with MCE
as CO and
Biomass burning experiments were performed under highly controlled
test conditions in a Fire Propagation Apparatus. Eight experiments
were performed in three different burning environments. In each case,
the samples were exposed to heat fluxes of 30 or 50
Emissions were observed to be dependent on three separate phases during small-scale wood burning: pyrolysis (phase 1), which produces mostly OA; flaming combustion (phase 2), when most of the rBC is produced; and smouldering-dominated combustion (phase 3), which emits primarily OA, although a luminous flame can still be seen. The emissions from any real-world burning event will comprise a variable combination of these phases taking place simultaneously within a mass of burning material. We have shown that the relative abundances of key gas and particulate components emitted during each particular combustion phase are similar across tests in different burning environments. However, the environment in which combustion takes place affects the duration of each phase, and smouldering behaviour is only observed when conditions exist to lower the flaming intensity.
Our observations show that the particulate emitted during pyrolysis and
smouldering is primarily OA, and that the OA in each of these burn phases is
chemically different. The composition of the former contains more reduced
hydrocarbons and the latter is characteristic of more oxidised hydrocarbons.
The ratio of
The differential chemical characteristics of these two phases are similar to those of two solid fuel organic aerosol factors identified by Young et al. (2015) in ambient air. This suggests that the differences in burn phases can contribute to the variability of ambient BBOA mass spectra and confirms that it is possible to infer burn conditions based on measurements of fresh emissions.
The emission factors for both aerosol species were consistent with results in the literature from experiments with similar modified combustion efficiencies. Within each experiment type and phase, emissions were reasonably consistent. Averaging over the entire period of combustion showed better correspondence between EFs and the MCE for OA than for rBC, which is consistent with results from previous experiments. However, the release of OA during pyrolysis and smouldering was not coupled to mass loss, which suggests that the correlation with MCE is best explained by the duration of the smouldering phase. The variability in rBC EFs could be related to a large increase in the rate of rBC emission above a certain threshold flame intensity.
The presentation of different chemical characteristics for three phases of biomass burning here provides a physical basis that could help to explain why there is such large variability in biomass burning emissions. This variability creates difficulties when producing bottom-up representations of emissions. If the differences in emissions associated with the pyrolysis and smouldering phases are better understood, however, and if further approaches to quantifying pyrolysis are developed, it could be possible to constrain emission estimates more completely. The development of real-world combustion differs from our experimental burns, which may affect the balance of emissions in the different phases. The yield of OA from the pyrolysis phase was largely dictated by the heat that reached the sample, with higher applied heat increasing the vaporisation of semi-volatile species. In the context of a wildfire, a number of variables could affect the rapidity and intensity of heating, and hence the pyrolysis yield. Examples of these include the fuel density, the fuel water content or the radiant heating of unburnt fuel adjacent to flaming combustion. The high radiant heat in the centre of a wildfire could show features of our high heat experiments, although this could be hindered by the fuel type or moisture content. Conditions are more constrained in a stove used for cooking or heating, where fuel is introduced in batches, but variations in fuel loading, moisture and type still result in more variation in emissions than has been seen here.
By being able to reproduce fire characteristics consistently between experiments and control the combustion environment, we have been able to explore and separate processes that form different aerosol components during its evolution. In doing so we have been able to relate a number of key aspects of the burn to real-world systems and to highlight why the systems behave in the way they do. Our work shows that whilst black carbon emission factors can be used robustly, caution needs to be used when applying OA emission factors averaged empirically over whole burns to other systems since the emission of OA is decoupled from the total mass loss. Further exploration of aerosol yields from such carefully controlled fires and relation to larger scale experiments may help to deliver greater constraints on the variability of particulate emissions in atmospheric systems.
Raw and processed data are archived at the University of Manchester and are available on request.
The supplement related to this article is available online at:
SLH, JCT, RH, JDA and HC designed the project; SLH, JCT, WTM, JDA, PIW and KS operated, calibrated and performed QA of instrument measurements; all authors contributed to the interpretation of data; SLH led the manuscript preparation, with JDA, HC, JCT, RH, DL, KS and CL contributing and critically reviewing.
The authors declare that they have no conflict of interest.
This work was supported by the UK Natural Environment Research Council (NERC) through the SAMBBA project (grant ref: NE/J010073/1) and the lead author was supported by the NERC Doctoral Training Programme (grant ref: NE/L002469/1). Albert Simoeni (now at Worcester Polytechnic Institute) is acknowledged for his input in the initial discussions and conception of this work. Edited by: James Roberts Reviewed by: three anonymous referees