GAMBUT field measurement of emissions from a tropical peatland fire experiment: from ignition to spread to suppression

Field site and peat soil characterisation

This controlled field-scale peatland fire experiment was conducted as part of the ‘1st GAMBUT Workshop: UK-Indonesia Collaboration for Mitigation of Peat Fires’. The objective of this workshop was to investigate the ignition, spread, emissions variability across the life cycle, and extinguishing of peatland fires. GAMBUT is the first study to fill the gap in the understanding of peat fire between laboratory and field scale, providing field evidence to formulate an effective and efficient mitigation response. The field fire experiment was specifically conducted from 19 August (Day 1) to 30 August (Day 12) 2018, in a secondary peat swamp area measuring 408 m² (34 m × 12 m) located in Rokan Hilir, Sumatra, Indonesia (Fig. 1). Climate history data suggests that the average temperature and humidity in the previous 5 years were 27.4 °C and 79.1%, and the mean daily rainfall in August is 6.5 mm, allowing for the investigation of peat fire emissions in a typical tropical environment (BMKG 2022; Santoso et al. 2022).

Fig. 1.

Map of Sumatra and southern peninsula of Malaysia showing the location of the experimental site at Rokan Hilir, Sumatra, Indonesia (1°36′17.1″N, 100°58′30.5″E) (Map Data, Google, 2023).

Site preparation was conducted prior to the ignition attempts. The peatland at the experimental site was manually divided into three separate and parallel plots, labelled as Plot 1, Plot 2, and Plot 3. There were live thick (>6 mm) fuels like palm trees, live thin fuels like ferns and sedges at the site. The original lush tropical plantation on each plot was manually cleared before the fire experiment. Specifically, dead thin fuel duff was kept intact for Plot 1, allowing the observation of its effect on slash-and-burn. Plot 2 and Plot 3 were left as bare peat ground with sparse palm tree roots (Fig. 2). Each plot had an interior dimension of 10 m × 10 m. Fire breaks, consisting of trenches 0.5 m wide and 0.5 m deep filled with sand, were constructed along the perimeter of each plot to prevent fires from spreading beyond the designated plot area.

Fig. 2.

Drone image showing the experimental site (a); Schematic of the experimental plots for GAMBUT peat fire experiment (b). Plot 1 was left with a shallow layer of surface litter vegetation thus was identified by green shading. The letters ‘N’ and ‘S’ in the plot name indicate the north and south sides of the plot, respectively.

Furthermore, each plot was divided into two sections: (1) the north side; and (2) the south side for different ignition and suppression attempts. Before the fire experiments, a thorough field site topology measurements were conducted, showing that there was an elevation difference of roughly 1 m between the north and south sides of the experimental site. A weather station was installed 10 m next to Plot 3 to monitor the atmospheric pressure, temperature, relative humidity, wind speed, and rain rate in the field. Further details about the experimental site in terms of the climate, topology, surface plantation, and surface treatment, are described in Santoso et al. (2022).

Representative peat sampling was conducted in situ prior to the ignition of the peatland to characterise the physicochemical properties of the soil, including moisture content (dry basis), wet bulk density, inorganic content (dry basis), and elemental analysis (the content of C/H/N). The samples were weighed to obtain the wet bulk density. The moisture content in dry basis was calculated by using the volumetric moisture content (VMC) measured from a soil moisture sensor probe (Delta-T Devices Ltd, England) and the wet bulk density. The inorganic content of the peat samples was derived from burning the sample in a furnace at 1000°C. Detailed calculation and determination of the physicochemical properties of the soil was elaborated in the twin paper of this study, in Santoso et al. (2022). This work follows the same metrics used in Santoso et al. (2022) in terms of the moisture and ash content in describing the soil properties, and in discussing the emission measurement results. For detailed locations of soil sampling, where PVC pipes were utilised to extract subterranean peat cores from nine sampling locations in each plot (0–40 cm depth), see Supplementary Fig. S1.

Ignition methods and emission measurements

The experiment commenced with ignition attempts on Day 1 (19 August). A charcoal ember ignition method employed in Pastor et al. (2017), and the traditional slash-and-burn approach commonly used in the oil palm plantation industry for peatlands conversion into plantation sites (Cochrane 2003), were applied in this study to ignite the peat soil, and dead ferns and sedges in different locations of the experimental plots and at different dates, respectively (Table 1). Specifically, a total of 9.3 kg of charcoal was used to produce the embers that were put in three pits (each with dimensions of 0.5 m × 0.2 m and 0.2 m deep) at P1S on Day 1 and Day 3 (Santoso et al. 2022). The charcoals were firstly ignited with gasoline and left to burn for 10 min, and then put into the ignition pit. Slash-and-burn, involving piling and igniting dry dead plantation materials such as tree branches, leaves, and litter, with dimensions of 8 m length × 1 m width × 0.5 m height, was conducted at P1N, P2N, and P3N on Day 5 and Day 7, respectively (Fig. 3). Details regarding the charcoal and slash-and-burn ignition protocols are provided in the twin paper of this study, in Santoso et al. (2022).

Fig. 3.

Schematic of locations of ember and slash-and-burn ignition attempts and an example of fire smoke plume measurement using a Fourier-transform infrared spectroscopy in the field.

Self-sustained smouldering was determined by visual observation of ground fire spread and fire size, and by examining the temperature profile of the soil using thermocouple readings and infra-red (IR) signatures (see Santoso et al. (2022) for details). Following successful ignition, the peatland started to smoulder from Day 2, slowly spreading towards unburned peatland areas and releasing fire smoke into the ambient air throughout the experiment.

Fig. 4 shows the timeline of the fire emission measurements from ignition to suppression. Specifically, a metal sheet was used in this study to cover a burning spot on Day 5 (23 August) when smouldering spread steadily, attempting to investigate the influence of limited oxygen supply on the emissions from the smouldering spread (Rein 2016). On Day 6, natural rainfall event occurred, with rain rate reaching an average of 5.6 mm h⁻¹ for more than 10 min. On Day 11 and Day 12, a water spray and a direct injection device were used to suppress and terminate the fires. Details regarding the suppression devices and protocols are illustrated in Santoso et al. (2022). In this work, a total of 40 different smoke plume measurement periods were undertaken for the characterisation of the life-cycle emissions including four fire type and stages observed in the field: ember ignition (EI), slash-and-burn (SB), smouldering spread (SS), and fire suppression (SP). In total, there were six ember ignition (EI1–EI6), 27 smouldering spread (SS1–SS27), four slash-and-burn (SB1–SB4), and three suppression (SP1–SP3) fire smoke measurements.

Fig. 4.

Timeline of the GAMBUT fire emission measurements from ignition to suppression. A total of 40 fire smoke plume measurements were conducted. Measurements from four fire types and stages, ember ignition (EI), smouldering spread (SS), slash-and-burn (SB), and suppression (SP) were included in this sketch.

Gas emissions from the fire smoke plumes were measured by using an open-path Fourier transform infrared spectroscopy (OP-FTIR). The OP-FTIR collects the real time spectra that contained absorption features for 13 target gases (CO₂, CO, CH₄, NH₃, acetylene (C₂H₂), ethylene (C₂H₄), ethane (C₂H₆), methanol (CH₃OH), formaldehyde (CH₂O), formic acid (HCOOH), hydrogen cyanide (HCN), acetic acid (CH₃COOH), and nitrous oxide (N₂O)). The OP-FTIR system used in this study consisted of a MIDAC Corporation M2000 Series FTIR spectrometer equipped with a Stirling-cooled mercury-cadmium-telluride detector and fitted with a MIDAC custom-built 76 mm Newtonian telescope. The spectrometer was mounted on an adjustable tripod to provide stable support for signal reception from a remotely located infrared source, which consisted of a 12-V silicon carbide glowbar operating at 1500 K fitted in front of a 20-cm diameter gold-plated collimator (Smith et al. 2018). The use of the OP-FTIR system for collecting biomass burning gas spectra has been detailed in previous studies (Wooster et al. 2011; Paton-Walsh et al. 2014; Smith et al. 2014).

In total, more than 9000 gas spectra were collected from these measurements. A forward modelling method combining the use of the Multiple Atmospheric Layer Transmission (MALT) program (Griffith 1996) and absorption line parameters adopted from the 2016 HITRAN transmission molecular absorption database (Gordon et al. 2017) were used to derive the path-averaged trace gas mole fractions. The spectral regions of the trace gases that contain the most sensitive features from (Paton-Walsh et al. 2014) were selected and fitted with synthetic spectra from MALT to retrieve the gas mole fractions for a known path length and meteorological parameters (atmospheric pressure and temperature) (Paton-Walsh et al. 2014; Smith et al. 2014, 2018). The derived gas mole fractions from the forward modelling method are expressed in μmol mol⁻¹ (ppm) and were found to have trustworthy accuracy (within 5%) and a small uncertainty of 3–5% for CO₂, CO, and CH₄ (Smith et al. 2011; Stockwell et al. 2016).

Fig. 5 shows four typical field emission measurements conducted during ember ignition, slash-and-burn, smouldering spread, and suppression attempts using the OP-FTIR. Each measurement period lasted between 10 and 110 min. Given the relatively short time slots for each plume measurement period (<2 h), compared with the broader scope of the fire evolution process (>2 weeks), each measurement period was deemed to be within a relatively steady fire stage (Hu et al. 2018b). Table 1 summarises the general information of the 40 peat fire smoke plume measurements.

Fig. 5.

Photographs showing a typical set-up of OP-FTIR measuring emissions from ember ignition (a, EI4), slash-and-burn (b, SB1), smouldering spread (c, SS23), and water spray suppression (d, SP3).

Emission factor and combustion efficiency quantification

Generally, there are two methods for calculating the EF of gaseous species from peat fires: (1) the mass loss approach; and (2) the carbon balance approach. The mass loss approach is mainly used in small-scale laboratory experiments where the mass loss rate of the peat is measured for calculating EF (Eqn 1) (Rein et al. 2009; Hu et al. 2018b):

where ṁı″ is the mass flux of the released species i (g s⁻¹ m⁻²), and ṁ″ is the mass loss rate (fuel consumption rate) of the dry peat (g s⁻¹ m⁻²).

The carbon balance approach is widely used in the literature to characterise peat fire emissions in the field (Stockwell et al. 2014, 2016; Smith et al. 2018). The carbon balance approach does not measure the mass loss of the peat, which is impractical in field measurements. instead, it requires information of the fuel carbon content. This approach assumes all carbon-containing emissions are measured (Eqn 2) (Ward and Radke 1993):

wwhere F_c is the carbon content of the fuel (%), MW_i is the molecular weight of species i (g mol⁻¹); 12 is the atomic mass of carbon (g mol⁻¹), C_i is the number of moles of species i (mol), and C_T is the total number of moles of carbon emitted (mol).

Both approaches have been verified in a laboratory-scale peat fire emission study (Hu et al. 2019). In this study, the carbon balance approach was used to derive the EFs of 11 targeted trace gas species. Ash-corrected carbon content of peat from the sampling locations across our site (Fig. S1) was calculated using Eqn 3:

where F_c-corrected is the ash-corrected carbon content of the fuel (%), F_s is the carbon content of the soil sample obtained from the elemental analysis (%), IC is the inorganic (ash) content of the soil sample in dry basis (%). For a summary on the carbon content, inorganic content, and ash-corrected carbon content of the fuel burnt across the experimental site, see Table S1.

In Eqn 2, the determination of C_i/C_T can be either calculated directly from the measured excess mole fractions from the OP-FTIR (Eqn 4) or using emission ratio (ER) with respect to a reference species (Eqn 5) (Paton-Walsh et al. 2014):

where Δ(i) and Δ(j) are the excess mole fractions of species i and j, respectively. The excess mole fraction is defined as the mole fraction measured (i) minus the mole fraction from the background (i)_background, NC_j is the number of carbon atoms in species j, and the sum is of all carbon-containing species emitted by the fire:

where ER_i/CO is the ER of species i to the reference species (CO in this work) (Eqn 6) (Smith et al. 2018):

When the amount of data used for regression is abundant, deriving the ER from the gradient of the linear best fit between species i and CO does not necessarily entail the knowledge of background mole fractions, yet introduces very low uncertainty (Wooster et al. 2011). Given the large amount (9000+) of spectra collected in this study and the negligible mole fraction of trace gases in the background, an ‘emission ratio to reference gas’ method (Paton-Walsh et al. 2014) was used to derive the EF for a particular species (except CO₂ and CO) (Eqn 7):

where MW_i is the molecular weight of species i (g mol⁻¹), MW_CO (28.01 g mol⁻¹) is the molecular weights of CO, and EF_CO is the EF of CO (g kg⁻¹).

In this work, EF_CO and EFCO2 are calculated by using a ‘summation method’ (Eqs. 8, 9) (Paton-Walsh et al. 2014; Smith et al. 2018). This method has a significant advantage providing accurate EF values but requires accurate background information to calculate the total excess amounts of each gas species (Paton-Walsh et al. 2014). As a result, background spectra collection using the OP-FTIR was carried out prior to each smoke plume measurement, ensuring a good knowledge of the mole fractions of trace gas species from the background.

where ΔCO₂ and ΔCO are the summed excess mole fractions of CO₂ and CO, respectively. Δj is the summed excess mole fractions of all carbon-containing species measured in this work (CO₂, CO, CH₄, C₂H₄, C₂H₆, CH₃OH, CH₃COOH, CH₂O, and HCN). These species account for ~>98% of all carbon emissions, omission of further carbonaceous species has been estimated to inflate the EFs by 1–2% (Yokelson et al. 2007).

In this work, modified combustion efficiency (MCE) from the four fire categories (ember ignition slash-and-burn, smouldering spread, and suppression) observed during the experiment was compared and discussed. MCE is defined as a proxy indicating the completeness of a combustion process (Ward and Hao 1991). The calculation of MCE is built on the use of the excess mole fractions of CO₂ (Δ[CO₂]) and CO (Δ[CO]) (Eqn 10) (Ward and Radke 1993; Yokelson et al. 1996). MCE has been used as a universal standard in the literature to determine the importance of flaming or smouldering in a fire (Christian et al. 2003; Stockwell et al. 2014; Urbanski 2014; Wilson et al. 2015):

Peat soil properties

Table 2 provides the properties including density, moisture content, inorganic content and the content of C/H/N of the peat soil sampled among the four plots (P1N, P1S, P2N, P3N) where the emission measurement campaign were carried out. Detailed physicochemical properties of the whole peatland site across depths investigated were shown in the twin paper of this work in Santoso et al. (2022). Significant variability was observed in terms of the physicochemical properties across the peatland site, P1S showed a lowest mean moisture content of 53.5 ± 18.7%, while P2N exhibited the highest mean moisture content (264.9 ± 136.2%) across depths measured. Wet bulk density ranged from 656 to 1483 kg m⁻³, with P1S (773.5 ± 79.6 kg m⁻³) and P1N (1326.8 ± 147.8 kg m⁻³) presenting the smallest and largest mean bulk density, respectively. In contrast, P1S and P1N exhibited the largest and smallest mean inorganic content across depths, respectively.

Elemental analysis result showed that P1S has the lowest mean carbon content (23.93%), while the largest peat carbon content value (31.3%) comes from P2N. In general, peat sampled from the experimental site has a lower carbon content than typical tropical peat (~55–60%) and a higher inorganic content compared with tropical peatland of 3 ± 1.96%, thus is deemed as degraded peat (Santoso et al. 2022). This degraded peat soil is commonly found in regions where the palm tree plantation industry is prevalent (Jauhiainen et al. 2016), which normally had undergone anthropogenic interference, such as drainage, logging, and agricultural conversion, as reported in previous studies (Page et al. 2011; Turetsky et al. 2015).

Gas mole fractions and emission ratio

Fig. 6 shows an example of transient path-averaged mole fractions of CO₂ and CO from a slash-and-burn attempt (SB3). It is evident that CO₂ and CO, two species predominantly generated from char oxidation (Rein et al. 2009), followed a similar evolution pattern. Owing to wind and natural convective processes in the field, there was significant variance in the gas mole fractions determined from the OP-FTIR throughout the measurements, with peak concentrations of CO₂ and CO reaching approximately 1100 and 200 ppm, respectively.

Fig. 6.

Example time series of path-averaged mole fractions CO₂ and CO from a slash-and-burn smoke plume (SB3).

To minimise the mixture of the adjacent smoke plumes, the path-averaged mole fractions of each measured gas species were retrieved from the OP-FTIR that was positioned directly above the actively burning peat. Furthermore, background mole fractions containing any possible emissions from the other adjacent burns were measured and subtracted from the calculated results. It is worth noting that some smoke plume measurements conducted during ember ignition and suppression exhibited gaps in the time series of gas mole fractions (e.g. EI2 and SS26, see Fig. S2), possibly attributable to periods of low signal-to-noise within the spectral window used for the retrieval of CO₂ and CO (Smith et al. 2018).

Fig. 7 shows a series of ER plots for investigated gas species against CO, for smoke plume SB3. Most species have a good correlation between species and excess mole fractions, indicating a well-mixed smoke plume in the field (Stockwell et al. 2016; Smith et al. 2018). Table S2 summaries ER values for investigated species against CO for all smoke plume measurements.

Fig. 7.

Example emission ratios (ER_i/CO) and the R² value of the measured gases in SB3, calculated from gradient of the linear best fit between y species (CH₄, C₂H₄, C₂H₆, H₂CO, CH₃OH, CH₃COOH, NH₃, HCN, and N₂O) and CO.

Emission factors and inter-plume emission factor variability

Summing up the excess mole fractions of all measured carbon-containing species, the EFs of CO₂ and CO were calculated using Eqns 8 and 9. Combining the EF of CO and the ER of the targeted species against CO, the EFs of the remaining nine gas species were calculated using Eqn 7. Table 3 summarises the EFs and the associated uncertainty of analysed gas species for all 40 fire smoke measurements (Paton-Walsh et al. 2014; Smith et al. 2018). In addition to the 11 species reported in this work, we attempted to retrieve C₂H₂ and HCOOH but these were either below the detection limit of the OP-FTIR, or had a very poor emission ratio correlation with CO (R² < 0.2); thus, these are not included.

Fig. 8 shows the individual EFs for CO₂, CO, CH₄ and NH₃, the four prominent peat fire gas species that are important for greenhouse gas accounting and air quality modelling. The figure shows the inter-plume variability across 40 smoke plume measurements. CO₂ exhibits large EF variability throughout the experiment. The percentage difference, defined as the difference between two values divided by the average of the two values expressed as a percentage (Smith et al. 2018), reached 90% between EI1 and SS26 where the maximum (2540.0 ± 254 g kg⁻¹) and the minimum (962.0 ± 96.2 g kg⁻¹) values of the CO₂ EF were derived, respectively. Substantial inter-plume variability in terms of the EF percentage difference was found for CO (154%), CH₄ (165%) and NH₃ (170%) throughout the experiment. The maximum EFs of CO (412.7 ± 41.2 g kg⁻¹), CH₄ (10.5 ± 1.4 g kg⁻¹) and NH₃ (8.5 ± 0.3 g kg⁻¹) were from smouldering spread smoke plumes, while the minimum EFs of those species were mostly obtained from slash-and-burn smoke plumes characterised by stoichiometric and complete flaming combustion (Rein 2016).

Fig. 8.

Path-averaged emission factor of CO₂ (a), CO (b), CH₄ (c), and NH₃ (d) for all 40 smoke plume measurements.

EF variability among fire/fuel types and fire stages

Fig. 9 compares the classified EFs of all detected species from four fire events (ember ignition, slash-and-burn, smouldering spread and suppression) observed in the field. Table 4 summarises the mean EFs of the smoke plumes from each fire category, and study-averaged EFs for all gas species measured. In general, gas emissions differ significantly among fuel types and fire categories (Rein 2016; Hu et al. 2019). Comparatively, ember ignition (EI1–EI6) has the largest CO₂ EFs (2446.5 ± 67.8 g kg⁻¹), averaging 56% higher than those from slash-and-burn, smouldering spread and suppression. This is mainly because the value of CO₂ EF is proportional to the carbon content of the fuel (Eqn 8). The charcoal ember consumed within this fire event has distinctively higher carbon content (78%) than those from peat (51.3–57.8%) and surface vegetation (55.0%) (Table S1), thus leading to a much higher EF value of CO₂. This finding verifies the important role of fuel composition affecting fire emissions (Yokelson et al. 1996; Akagi et al. 2011; Smith et al. 2018; Hu et al. 2018a).

Fig. 9.

Emission factor of CO₂ (a), CO (b), CH₄ (c), C₂H₄ (d), C₂H₆ (e), CH₃OH (f), CH₂O (g), CH₃COOH (h), NH₃ (i), N₂O (j), and HCN (k) from different fire categories. Each box represents groups of emission factor data through their lower quartile (25th percentile), the median (50th percentile) and upper quartile (75th percentile). The mean values are given as solid squares. The error bars show the range of the emission factors in each group.

The surface vegetation burnt from slash-and-burn (SB1–SB4) and peat burnt from smouldering spread peat (SS1–SS27) have similar carbon content. In comparison, slash-and-burn has 6% higher CO₂ EF (1693.4 ± 98.4 g kg⁻¹) but ~40 and ~66% lower CO EF (127 ± 73.3 g kg⁻¹) and HCN EF (1.2 ± 0.46 g kg⁻¹) than those from smouldering spread. This is attributed to the fact that slash-and-burn is dominated by flaming combustion, which has a higher combustion efficiency and a higher conversion ratio of the carbon from the fuel to complete combustion products (e.g. CO₂) than smouldering (Rein 2016; Hu et al. 2018a). The EFs of CH₃OH (1.67 ± 0.95 g kg⁻¹), CH₂O (3.2 ± 1.2 g kg⁻¹), and CH₃COOH (5.2 ± 5.4 g kg⁻¹) from slash-and-burn stayed close to their corresponding EF values from the burning of crop residue reported in Akagi et al. (2011). In this field measurement, smouldering spread showed a ~300% times higher EF value (3.57 ± 2.03 g kg⁻¹) than flaming slash-and-burn for NH₃, a typical incomplete combustion product and a critical nitrogenous species for forming haze (Plautz 2018). This agrees with the findings from the literature that EFs of NH₃ from smouldering peat were significantly higher than flaming biomass burning (Akagi et al. 2011; Hu et al. 2019).

Plume-averaged EFs of CO₂ (1591 ± 243 g kg⁻¹), CO (206.4 ± 85 g kg⁻¹), CH₂O (1.2 ± 0.5 g kg⁻¹), CH₃OH (2.08 ± 1.24 g kg⁻¹), CH₃COOH (4.7 ± 2.0 g kg⁻¹), NH₃ (3.57 ± 2.03 g kg⁻¹), and HCN (3.4 ± 1.7 g kg⁻¹) from smouldering spread at this degraded peatland stayed within the range of their EFs reported in the literature (Huijnen et al. 2016; Stockwell et al. 2016; Smith et al. 2018). EFs CH₄ (4.3 ± 2.4 g kg⁻¹) and C₂H₄ (0.4 ± 0.16 g kg⁻¹) measured during smouldering spread stayed at the low end of those reported in peer studies, while the EF of C₂H₆ (5.6 ± 2.6 g kg⁻¹) stayed at the high end (Hu et al. 2018a). The inclusion of these new EFs in the EF inventory could contribute to a better understanding of emissions and their inherent variability for degraded tropical peatland fires.

Gas emissions from suppression attempts (SP1–SP3) were investigated for the first time in this experiment. Significant variability was found for EFs for most species. For example, the EF for CO from the suppression stage stayed between 68.5 and 391.4 g kg⁻¹, while CH₄ and NH₃ EFs ranged between 2.9–9.3 and 1.0–8.3 g kg⁻¹, respectively. The large variability of the EFs seen at this fire stage is possibly caused by the limited amount of smoke plume measured (n = 3) at different peatland locations (P1N, P1S, P2N) as well as the different methods (water spray and injection) and water usage used in each fire suppression attempt (Santoso et al. 2022), likely affecting combustion efficiency and emissions in different ways.

Independent sample t-tests were conducted in this study to statistically examine whether EFs of the four most prominent gas species (CO₂, CO, CH₄, NH₃) from either ember ignition, slash-and-burn, smouldering spread and suppression are significantly different from the others. Specifically, six t-tests (two-tailed) with a significance level of 0.05 (a = 0.05) were carried out for: (1) ember ignition vs slash-and-burn; (2) ember ignition vs smouldering spread; (3) ember ignition vs suppression; (4) slash-and-burn vs smouldering spread; (5) slash-and-burn vs suppression; and (6) smouldering spread vs suppression (Table 5).

CO₂ showed a significantly different EFs (P < 0.001) from ember ignition (n = 6) than from slash-and-burn (n = 4), or smouldering spread (n = 27), or suppression (n = 3). CO₂ EFs from suppression also showed significantly different values than those from slash-and-burn (P < 0.001), and from smouldering spread (P = 0.0077). However, CO₂ failed to differ EFs between slash-and-burn, and smouldering spread (P = 0.262). Compared with CO₂, CO only performed well in distinguishing ember ignition (n = 6) from slash-and-burn (n = 4) with P = 0.0385. CH₄ failed to differ any fire types or fire stages with all P-values staying above 0.05. For NH₃, a potential gas signature for smouldering peat proposed in (Hu et al. 2018b), performed well in distinguishing ember ignition (n = 4) from slash-and-burn (n = 4), and smouldering spread (n = 27) from slash-and-burn, with both P-values <0.0001. More field fire emission measurements are needed for improving the statistical performance of single gas species in differentiating fire types and stages.

Influence of soil properties on smouldering fire emissions

The percentage difference of EFs of CO₂, CO, CH₄, and NH₃ from the 27 smouldering spread smoke plume measurements stayed at 51.8, 86, 90 and 88%, respectively. In laboratory studies, soil properties (e.g. moisture content, inorganic content and bulk density) have been shown to affect the smouldering fire dynamics (Huang et al. 2016; Huang and Rein 2017; Christensen et al. 2019; Hu et al. 2019, 2020). In this field experiment, the influence of these soil variables on smouldering fire emissions were examined.

Averaging the soil sampled across depths in the field, peat from P1N was found to have a 225% higher moisture content (173.9 ± 22% in dry basis) than from P1S. Fig. 10 shows the relationships between the peat moisture and the mean EFs of CO₂, CO, CH₄, and NH₃, from plots P1N, P1S, and P2N where fire emissions from the smouldering spread stage were measured. The mean EF of CO₂ (1356.5 ± 186.2 g kg⁻¹) from smouldering smoke plumes SS13–SS14, SS16, SS21–SS24, and SS26–SS27 measured at P1N were 23% lower than the mean EF value of CO₂ from SS1–SS10, SS15, SS17–SS18, and SS20 measured at P1S.

Fig. 10.

Relationships between the emission factor of CO₂, CO, CH₄, and NH₃ and mean moisture content of the plots P1S, P1N and P2N where fire emissions from smouldering spread were measured. Smoke plumes influenced by limited oxygen supply (SS11–SS12), and rainfall (SS19) were excluded from the results. Only one measurement of smouldering spread fire emissions was carried out in P2N, the data was included in this figure for comparison.

In contrast, the mean EF of CO (225.7 ± 82.5 g kg⁻¹) and CH₄ (6.0 ± 2.45 g kg⁻¹) from P1N was 48.8 and 50.0% higher than the mean EF of CO and CH₄ from P1S, respectively. These observed changes in EF with peatland moisture content is in accordance with the findings from laboratory experiments; wet peat has a lower EF of CO₂ but a higher EF of CO as moisture content decreases combustion intensities (Hu et al. 2019). However, it is worth noting that transient moisture of the peat is susceptible to ambient humidity and rainfall, which vary significantly in the field. More fire emission measurements under diverse moisture conditions are needed to improve the understanding of the influence of moisture on smouldering fire emissions.

Fig. 11 shows the relationships between mean gaseous EFs and the density and inorganic content from P1S, P1N, and P2N. In general, increasing in density leads to a decrease in CO₂, but an increase in the EFs of CO, CH₄, and NH₃. Conversely, when the inorganic content of the peat soil increase, an increase in CO₂ and decreases in CO, CH₄, and NH₃ were observed. Specifically, EFs of CH₄ from P1N (6.0 ± 2.5 g kg⁻¹) with 72% higher peat bulk density were found to be 99% larger than the EFs of CH₄ from P1S. This echoes with the finding from a peat fire emission field study where a strong positive correlation was found between the peat substrate bulk densities with CH₄ EFs (Smith et al. 2018).

Fig. 11.

Relationships between the emission factor of CO₂, CO, CH₄, and NH₃ and mean density (a) and inorganic contents (b) of the plots P1S, P1N, and P2N where fire emissions from smouldering spread were measured. Smoke plumes influenced by limited oxygen supply (SS11–SS12), and rainfall (SS19) were excluded from the results. Only one measurement of smouldering spread fire emissions was carried out in P2N, the data was included in this figure for comparison.

In essence, a higher bulk density can be resulted from either a higher degree of packing, or a higher inorganic content. Higher bulk density fuel with a higher inorganic content may have an improved oxygen supply and thermal conductivity, resulting from the inclusion of smaller and thermally conductive mineral particles in the peat fuel bed. In contrast, higher bulk density peat soils resulting from a tighter structure entails a slower heat loss and a limited oxygen supply inside the peat bed, encouraging the formation of incomplete combustion species (e.g. CH₄) from peat pyrolysis (Wijedasa et al. 2017; Smith et al. 2018; Hu et al. 2020; Cui 2022). These findings suggest that it is important to map the peatland heterogeneity (e.g. moisture content, inorganic content and bulk density) for a better understanding of spatially variable EFs.

Influence of field meteorological conditions on smouldering fire emissions

In the field, peatland fires are subjected to wind with varying speeds and directions. In SS11, a burning spot at P1S was partly covered by a metal sheet for 20 min (Table 1). When removing the metal sheet (SS12), the accumulated emissions from the spot was released into the ambient air, resulting in a 27% decrease in CO₂ EF but a sharp 111, 239, and 263% increase of EFs for CO, CH₄, and NH₃, respectively. The field measurement result showed above indicates that changes in oxygen supply and heat loss at smouldering spots could significantly influence fire EFs. Reversely to the metal sheet effect (limited oxygen supply and heat loss from the smouldering spots) demonstrated from SS11 to SS12, strong wind or wind gusts could lead to an enhanced oxygen supply and heat loss at certain smouldering spots in situ, potentially introducing inter-plume EF variability (Rein 2016).

Among the 40 smoke plumes measured throughout the experiment, SS19 was conducted during a heavy rain event with an average rain rate of 5.6 mm h⁻¹. Compared to SS16 and SS21 (two rain-free measurements at the same location (P1N)), EF of CO₂ from SS19 increased 6.7 ± 1.3%, reaching 1534.2 ± 153 g kg⁻¹. In contrast, a notable decrease in the values of EFs for CO (−17.2 ± 3.4%), CH₄ (−10.8 ± 3.8%), C₂H₆ (−25.3 ± 20%), CH₄O (−23.8 ± 2.6%), NH₃ (−40.3 ± 6.7%), HCN (−53.7 ± 10.9%), and N₂O (−62.5 ± 36.5%) were found from this SS19 smoke plume influenced by the rain.

Without rainfall, species like CO₂ and CO are mainly generated from the oxidation of ‘dry char’ (Rein et al. 2009; Rein 2016). However, the rainfall could convert the ‘dry char’ into a ‘wet char’, influencing the composition and concentrations of the fire smoke and thus leading to the changes of the EFs for CO₂, CO, CH₄, and C₂H₆ observed from SS19. The sharp decrease of EFs for gas species like CH₂O, NH₃, HCN, and N₂O could be partly attributed to the fact that they are soluble in water (rainfall). It is worth noting that the influence of the rain on fire EFs shown here is based on a single rain smoke plume measurement. However, this first investigation of the influence of rainfall on fire emissions opens up opportunities for further study.

Emissions from rain (SS19) and suppression attempts (SP1–SP3) were both influenced by water. Comparing with the EFs from SS19 with rain effect, the averaged EFs from SP1–SP3 measurements presented a mild decrease (7.34%) in the EF value of CO₂, but sharp increases in CO (36.0%), CH₄ (75.8%), C₂H₆ (29.0%), NH₃ (82.6%), and HCN (68.4%). However, owing to the limited measurements for rain (n = 1) and suppression smoke plumes (n = 3), the accurate roles of diverse forms of water (rain, water spay and water injection) in influencing peat fire emissions remain unclear. Further investigations are needed to reveal the mechanisms leading to diverse EFs influenced by different forms and intensities of water. However, this work presents the first field measurements of peat fire EFs during rainfall/suppression, contributing important primary baseline data.

Modified combustion efficiency (MCE) and fire regimes

In this work, the plume-averaged MCE from all measurements were classified and compared to the fire categories observed during the experiment (Fig. 12). It was found that MCE failed to differentiate the four fire categories that differ fundamentally in terms of the combustion dynamics. For example, the plume-averaged MCE from ember ignition (0.88 ± 0.01) exceeded the ‘smouldering MCE range’ (between 0.75 and 0.84, or between 0.65 and 0.85) defined in the literature (Akagi et al. 2011; Stockwell et al. 2016). However, only smouldering combustion was observed throughout the ember ignition attempts. In addition, slash-and-burn, a fire type dominated by flaming combustion, presents a plume-averaged MCE of 0.89 ± 0.06, staying roughly at the same MCE level with smouldering ember ignition.

Fig. 12.

Plume-averaged modified combustion efficiency from the smoke measurements of ember ignition, slash-and-burn, smouldering, and suppression. Each box represents groups of data through their lower quartile (25th percentile), the median (50th percentile), upper quartile (75th percentile). The mean values are given as solid squares. The error bars show the range of the modified combustion efficiency in each group.

In the literature, an MCE of 0.9 suggests roughly equal amounts of flaming and smouldering (Akagi et al. 2011). However, the plume-averaged MCE for SB4, a plume measurement that mixed partial flaming and smouldering smoke, stayed at 0.8, a typical MCE value for pure smouldering (Akagi et al. 2011). The maximum MCE (0.956) found in a smouldering smoke plume (SS6) stayed closely to the ‘pure flaming MCE’ (0.99) defined in (Stockwell et al. 2016). In addition, individual MCE from the three suppression smoke plumes ranged from 0.71 (SP1) to 0.93 (SP3), spanning over the whole literature MCE range for smouldering (Akagi et al. 2011; Stockwell et al. 2016). Given the large range of the MCE throughout the experiment (0.66–0.96) and the large contradictions seen between the ‘MCE-indicated fire regimes’ and real fire regimes observed in the field, we conclude that the value of MCE is highly sensitive to complex field variables including fuel heterogeneity and weather conditions (Hu et al. 2018a).

In a wildfire event, various biomass fuel (e.g. peat and surface vegetation) undergo chemical decomposition by heating from ignition sources (e.g. slash-and-burn, natural lightning or arson), generating gaseous (pyrolysate) and solid (char) products (Rein 2013). Fundamentally, the dominant mode of a fire regime (smouldering or flaming) is dictated when chemical species is oxidised; if the oxidation takes place in the char, smouldering dominants, while if the oxidation happens in the gas phase pyrolysate then flaming combustion dominates (Rein 2016).

In this experiment, fire emissions are affected by the peatland heterogeneity and complex weather conditions. These field (‘real world’) variables could have substantial influence on the transient CO₂ and CO emissions and in turn, lead to significant variations of the associated values of MCE (see Fig. S3 for the transient CO₂, CO, and MCE for SS26 as an example). MCE greatly simplifies the complex fire dynamics that drive changes in fire emissions, and fails to incorporate the fundamental difference between flaming and smouldering combustion (Hu et al. 2018b). As a result, large uncertainties could be introduced when using a single MCE value to determine the regimes of a wildland fire.

It is worth noting that comparing to natural smouldering peatland fires spreading thousands of km² and releasing emissions over weeks or even months, the smouldering area of the GAMBUT fire experiment (408 m²), as well as the number of the smoke plume measurements (n = 40) conducted in this research remain limited. Furthermore, uncertainties exist in terms of the accuracy of the EFs and MCE reported from the FTIR measurements of plume emissions subjected to wind gusts and potential mixing of emissions from adjacent burns. However, this field fire research provides a framework for an advanced understanding of temporally variable EFs and emissions inventories from fires on degraded tropical peatlands.