Comparing gas composition from fast pyrolysis of live foliage measured in bench-scale and fire-scale experiments

Introduction

Wildland fire is a complex phenomenon involving many chemical and physical processes at various length scales. Moldoveanu (2021) categorised wildland fire as burning which is a process that includes both pyrolysis and combustion. The two processes are not clearly separable in a natural setting since the products of one process are linked to and can influence the other process. During pyrolysis, a dead, solid wildland fuel composed of cellulose, hemicellulose, lignin, water and trace elements is heated and breaks down into constituent parts consisting of gases, tars and a solid material called char (Shafizadeh and Fu 1973; Shafizadeh 1982; Di Blasi 2008; Neves et al. 2011). Living wildland fuels additionally contain plant metabolites (Jolly et al. 2012, Matt et al. 2020). While Shafizadeh and others examined pyrolysis in non-oxidising and oxidising environments, Di Blasi’s modelling review (2008) limited pyrolysis to a non-oxidising environment. In a wildland fire, pyrolysis occurs in the presence of oxygen and combustion products (Leroy et al. 2006). For porous plant materials, oxygen and water vapour can modify the pyrolysis process, either as catalysts to the celluloses or char oxygen chemisorption (DeGroot and Shafizadeh 1983; Moldoveanu 2021). Smouldering combustion produces pyrolysis products that are a result of a low heating rate in a relatively oxygen-deficient environment (Frandsen 1991; McKenzie et al. 1995; Carvalho et al. 2002; Torero et al. 2020). During flaming combustion, gaseous pyrolysis products generated by higher heating rates can react with oxygen releasing energy (heat) and a large assortment of gaseous and solid chemical compounds and solid products (char) which can also react further with oxygen via the glowing combustion process.

An extensive body of knowledge about pyrolysis of wildland fuels in both non-oxidising and oxidising environments exists. Numerous compounds have been identified from pyrolysis of wood and foliage. For example, 213, 224, and 326 compounds have been previously listed (Goos 1952; Soltes and Elder 1981; Weise et al. 2022b), respectively, and other recent work identified potentially hundreds of compounds (Moore et al. 2021). The composition of pyrolysis products is dependent on heating rate and temperature (e.g. Shafizadeh 1982; Safdari et al. 2019). Temperatures below 300°C favour char, H₂O, CO and CO₂ production, intermediate temperatures (300–500°C) favour production of tar containing many different compounds and higher temperatures (>500°C) produce a mixture of low molecular weight gaseous products (Shafizadeh 1982). Safdari et al., (2019) found that fast pyrolysis of foliage from live fuels resulted in higher tar and lower char yield when compared to slow pyrolysis that was like thermogravimetric analysis (TGA). Gas yield between the two pyrolysis types was not statistically different. In a study of the O₂ effect on cellulose degradation in smouldering combustion, the difference in pyrolysis rate at a heating rate of 0.25°C s⁻¹ under nitrogen and air atmospheres disappeared at 310°C (Shafizadeh and Bradbury 1979).

Much of the work describing pyrolysis kinetics has occurred in benchtop-scale TGA-type experiments at low or high heating rates and the results have been assumed to be applicable to wildland fires (Philpot 1970; Dimitrakopoulos 2001). However, TGA experiments typically only use slow heating rates (<1°C s⁻¹) whereas measured heating rates of 75–5000°C s⁻¹ in wildland fires have been reported (Butler et al. 2004; Tachajapong et al. 2008). TGA experiments are applicable to conditions under which smouldering combustion occurs and have been critical to defining processes associated with smouldering combustion (Torero et al. 2020). In addition, TGA and other similar methods most often grind and dry fuels to produce uniform samples to yield more consistent, least adulterated results, but this method raises the question about loss of organic material from the drying and grinding process and the effect of fuel particle shape. While most TGA experiments occur in inert atmospheres, some have been conducted in normal atmospheres with the intent to be more applicable to the wildland fire setting (Hillado 1977; Dimitrakopoulos 2001). Conditions under which pyrolysis data from inert atmospheres can be extended to pyrolysis under an oxidising environment have been identified: (1) oxygen cannot reach the surface of a pyrolysing particle or (2) the type of environment (inert, oxidising) does not affect the kinetics and the pyrolysis mechanisms (Senneca et al. 2004); however, it is unknown if these conditions are implicitly assumed to be met in pyrolysis models used in wildland fire. A quick review of the literature reveals that the type of environment may or may not affect the kinetics of pyrolysis.

Tihay et al. (2009a) described three general approaches to modelling gas phase combustion in physical fire models: using global rates and thermodynamic parameters, using reduced reaction mechanisms, and using CO burning in air. In a review of wildland fire models developed between 1990 and 2009, Sullivan (2009a, 2009b, 2009c) described how several models implemented pyrolysis and combustion. Simplified reactions and chemical kinetics, often involving char and gases, were used in the physical models because of the high computational demand needed to implement detailed chemistry. Many of the physics-based fire models used a variety of mixtures of key pyrolysis products including H₂, CO, CH₄, and C₂H₆ (Grishin 1997; Grishin and Perminov 1998; Morvan and Dupuy 2001; Zhou and Mahalingam 2001; Mell et al. 2009; Clark et al. 2010; Borujerdi et al. 2022) and solid char (Grishin 1997; Grishin and Perminov 1998; Morvan and Dupuy 2001; Zhou and Mahalingam 2001; Mell et al. 2009; Clark et al. 2010; Borujerdi et al. 2022) and the process of char oxidation. The assumed mixture of gases used in these models was typically chosen based on a simplified pyrolysis reaction model, with reaction kinetic parameters provided by thermogravimetric analysis (TGA) experiments of wood and other wildland fuels performed in either a non-oxidising or oxidising environment.

QUIC-Fire is an example of a complex physical model that is being simplified for ‘field’ use (Linn et al. 2020). The chemistry has been simplified to a single equation representing the burning of wood with oxygen. In the United States, today’s widely used operational fire behaviour model was based on data from homogenous beds of dead fuels and associated theory (Rothermel 1972). The chemical aspects of combustion were assumed to occur at much faster rates than heat transfer and the only chemical aspect of this model related to fuel heat content (kJ kg⁻¹). Thus, heat content has been determined for many common wildland fuels and used in the basic calculation of fire spread (Hough 1969; Countryman and Philpot 1970; Rothermel 1972; Countryman 1982; Susott 1982a, 1982b; Rogers et al. 1986).

While TGA and fast pyrolysis techniques have been applied to a variety of lignocellulosic fuels, fast pyrolysis techniques have seldom been applied to wildland fuels, particularly the intact foliar portion which is typically the most readily ignited component in a wildland fuel complex. Similarly, the results from the many lab-based, tightly controlled experiments have seldom been compared to measurements made in wildland fires which are more complex to ascertain similarity between lab and field conditions. The applicability of the assumptions and simplifications made in the models to the actual conditions under which a wildland fire pyrolyses unburned fuels is largely undescribed. We found no literature supporting the assumption that bench-scale experiments are applicable to the wildland fire setting (Weise et al. 2022a).

It is well known that primary pyrolysis products that enter a hot environment may further react even in the absence of O₂. One pathway is for the primary tars to crack into lower molecular weight gas species. Another pathway is for the aromatic tars to polymerise, possibly with the help of some of the light gases such as acetylene and ethylene, forming polyaromatic hydrocarbons (PAH) and eventually forming soot. The polymerisation and condensation reactions release H₂ in the process. Pathways in the presence of O₂ in a fire include combustion, gasification, and water-gas shift reactions, depending on the mixing, local temperature, and local species concentrations. Comparison of the oxidative and non-oxidative pyrolysis results can help shed light on which processes are important in the flames studied.

In a project (Weise et al. 2022a) designed to improve the understanding and modelling capability of pyrolysis in physics-based fire spread models, three goals were identified: (1) provide more detailed description of pyrolysis and the evolution of its products for a greater variety of southern United States fuels than is currently known, (2) determine how convective and radiative heat transfer from flames to live fuel particles influenced pyrolysis and ignition at laboratory and field scales and (3) gain more detailed insight into pyrolysis, combustion and heat transfer processes in wildland fire spread through the use of high-fidelity physics-based models. These goals were achieved by (1) characterising the physical, chemical, compositional and spatial structure of wildland fuels (Safdari et al. 2018; Hudak et al. 2020; Matt et al. 2020; Herzog et al. 2022), (2) characterising pyrolysis products by measurement of a variety of live and dead foliar fuels in laboratory and small-scale field experiments (Safdari et al. 2018; Amini et al. 2019a; Scharko et al. 2019a, 2019b; Phillips et al. 2020; Banach et al. 2021; Moore et al. 2021), (3) determining the effects of convective and radiant heat transfer on pyrolysis (Safdari et al. 2019, 2020; Weise et al. 2022b) and (4) performing high-fidelity physics-based modelling of pyrolysis for bench-scale and wind tunnel experiments (Yashwanth et al. 2015, 2016; Shotorban et al. 2018; Borujerdi et al. 2019, 2020, 2022; Borujerdi and Shotorban 2022). Pyrolysis gases were generated in non-oxidising environments from individual leaves at bench-scale by slow heating in a pyrolyser (Hillier et al. 2013) and fast heating in a flat-flame burner at heating rates and temperatures typical of wildland fires (approximately 100°C s⁻¹, 100 kW m⁻² s⁻¹ and 765°C) (Safdari et al. 2018). Pyrolysis gases from fuel beds of live and dead fuels were measured in a wind tunnel and in the field in small prescribed burns at Ft. Jackson, SC by capturing gases in canisters (Weise et al. 2022c) or in real-time using nonintrusive Fourier-transform infrared (FTIR) spectroscopy (Scharko et al. 2019a; Banach et al. 2021).

At the bench scale, heating rate/mode, moisture status and plant species affected the composition and yield of pyrolysis products with higher heating rates yielding relatively more tars (Safdari et al. 2019; Weise et al. 2022b). In the oxidising (normal atmosphere) environment of the wind tunnel and small prescribed burns, burning phase (pyrolysis, flaming combustion) affected the gas composition. Log-ratios of measured gases with CO₂ were larger in the pyrolysis samples indicating relatively more of these gases in the pyrolysis samples. The presence of live plants affected the gas composition. The relative composition of pyrolysis gases was affected by the fire location (wind tunnel, field) (Weise et al. 2022c). Comparison of the FTIR and gas chromatography (GC)/flame ionization detector (FID) gas compositions found that trace gas composition differed between methods but each method yielded comparable description of the primary fuel gases (CO, CH₄) (Weise et al. 2023). This paper presents the comparison of the pyrolysis data derived from these experiments to test the assumption that TGA-like and fast pyrolysis are representative of wildland fire conditions.

Methods

A brief description of each of the three experiments follows. The interested reader is referred to the referenced papers for more details. All papers from the Forest Service are available through the www.fs.usda.gov/research/treesearch document repository and the data sets are available through the Wildland Fire Science Initiative (WFSI) Data Portal (Weise et al. 2024a, 2024b, 2024c).

Bench-scale non-oxidative pyrolysis experiments

Bench-scale experiments at Brigham Young University (BYU) used a flat-flame burner (FFB) with and without a radiant panel to produce a hot, oxygen-free post flame environment to heat single species samples, typically leaves, at high heating rates (180 and 195°C s⁻¹) in a non-oxidising environment (Safdari et al. 2020). The FFB setup (Fig. 1a) yielded some permanent gases and many tars comprising the measured pyrolysis products. These gas samples and all others described below are time-integrated samples. While the BYU FFB setup has been used previously to measure mass loss associated with ignition of live fuels (e.g. Fletcher et al. 2007) and mass loss recorded during the current experiments (Safdari 2018), the composition of the gas samples was determined off-line. Foliage from 15 different plant species was heated in the FFB. Tars, which can undergo secondary pyrolysis producing additional simpler pyrolysis products, and kinetic modelling, were the primary foci of this set of experiments (Safdari et al. 2018, 2019, 2020; Amini et al. 2019a, 2019b, 2021). The yield and concentrations of over 100 tar compounds were determined using a gas chromatograph equipped with a mass spectrometer (GC/MS). A gas chromatograph equipped with a thermal conductivity detector (GC/TCD) was used for permanent (light or non-condensable at ambient conditions) gases. The permanent gases consisted of CO, CO₂, CH₄ and H₂. Molecular mass of the identified gases ranged from 2 to 250 amu. As the primary focus of the experiment was to measure tars and other secondary pyrolysis products, GC/MS was selected for detection. To eliminate the effect of heating rate on composition in this analysis, only data resulting from convective or radiant/convective high heating rates, denoted Conv and RadConv, respectively, were used from the bench-scale experiments. In an on-going companion bench-scale study at the USDA Forest Service Forest Products Laboratory, a cone calorimeter has been used to heat leaf samples at irradiance of 35 kW m⁻² to achieve fast pyrolysis (10°C s⁻¹) in air for both unpiloted and piloted conditions; however, those data are not yet published and they do not provide as detailed a description of the composition of permanent gases and tars as the BYU data. The pyrolysis gas composition determined at BYU was used in the physics-based modelling of leaf scale flames (Borujerdi et al. 2022). While not pertinent to the present paper, related measurements and modelling to estimate kinetic parameters for pyrolysis of the BYU leaf samples and longleaf pine needles based on slow pyrolysis are available (Amini et al. 2019a, 2021; McGrattan et al. 2023, sec. 12.9.4).

Fig. 1.

Experimental setups to determine composition of fast pyrolysis gases in non-oxidising (a) and oxidising (b–d) environments. (a) Flat-flame burner setup used to heat southern U.S. wildland fuels in an O₂-free (fuel rich) post-flame environment. (b) canister sampling for flaming (F) and pyrolysis gases (P) in a wind tunnel where the fire's flame heated fuels. (c) FTIR and GC/FID sampling setup in the wind tunnel. (d) canister sampling for FTIR and GC/FID in the field.

Results

The combined data set consisted of 306 observations (bench = 174, wind tunnel = 107, field = 25). In the non-oxidising bench scale experiments at BYU, all six gases were measured. In the oxidising environment experiments in the wind tunnel and in the field, the GC/FID analysis of the canisters did not measure phenol. Infrared spectroscopy is unable to detect homonuclear diatomic molecules of which H₂ is one. These unmeasured/undetected gases were assigned below detection limit (BDL) values in these samples using the multLN procedure as described above resulting in a BDL percentage of 34% for phenol (105/306) and 8.8% for H₂ (27/306).

The sample mean compositions changed notably between the non-oxidative pyrolysis (bench) samples and the oxidative (wind tunnel, field) samples (Table 2 top). Except for CO₂, the mean values listed in Table 2 were higher in the bench measurements than in the wind tunnel or field experiments. Because of dilution factors and familiarity with gas ratios in syngas produced by pyrolysis, we examined the untransformed ratios of H₂/CO or H₂/CO₂ as an indicator of the validity of the bench measurements. These ratios were negligible in the wind tunnel and field experiments. In the bench experiments the ratios were 0.38 and 1.31, respectively, which was similar to syngas ratios produced from coal and biomass (Di Blasi 1999; Grieco 2011).

Table 2 (bottom) shows the percentage contributions of the individual gases to the metric variance of the composition. The contribution of CO was relatively small and comparable between the locations; those percentages of CO₂ and CH₄ were similarly low for the wind tunnel and field. While the contribution for H₂ was relatively small in the bench-scale measurements, it was more than half of the variability in the wind tunnel and field. Similarly, phenol (C₆H₆O) accounted for somewhat more than half of the variability at the bench-scale. In addition to phenol, 18 other phenolic compounds were identified in the tars collected in the bench-scale experiments (Safdari et al. 2018) and both plant species and heating mode/rate affected the relative amounts of these compounds (Weise et al. 2022b). The variability of the individual phenolic compounds ranged up to 2% of the total variance in these experiments. In the wind tunnel measurements, the GC/FID did not measure phenol which was assigned a BDL value described previously, resulting in a smaller contribution to total variance. In the wind tunnel, a heated line was used to keep sampled gases from condensing prior to injection in the White cell used with the FTIR and a HEPA (high efficiency particulate air) filter was used to remove any condensed tars and char particles (Banach et al. 2021). In the field, gas samples were collected in SUMMA canisters; gas samples were extracted from the unheated canisters within 24 h via a 70°C heated line and injected into the White cell (Scharko et al. 2019b). This temperature is not sufficient to keep tars from condensing. These sampling processes likely decreased phenol’s relative amount and contribution to total variance in the wind tunnel and field locations.

In the compositional PCA biplot of the gas composition (Fig. 2), the first principal component (PC1) accounted for 60.7% of the variation and the second principal component (PC2) accounted for an additional 33.1% for a total of 93.8%. Several characteristics of the data can be discerned from Fig. 2. The length of rays approximates the standard deviations of the clr-transformed gases. These were small for CO and CH₄ relative to the other gases in the composition, which is consistent with their overall low contribution to the total variability of the composition as reported in Table 2. Having the rays for CH₄, CO and CO₂ pointing in the same direction and close to each other indicated that these gases were highly proportional in concentration. Proportionality is the CoDA analog to correlation (Egozcue et al. 2018, Lovell et al. 2015). Contrarily, the ray for phenol pointing in the opposite direction, and exhibiting long links between the corresponding arrowheads, suggested a negligible association with CO₂, CH₄ and CO concentrations. Similar pattern was observed between H₂ and the Other gases. These results showed that the bench-scale observations (non-oxidative environment) were tightly clustered and distinct from the wind tunnel and field observations, particularly along the direction defining the contrast between phenol and CO₂ concentrations, with bench-scale observations presenting higher concentrations of phenol and lower concentrations of CO₂, comparatively. This result is consistent with the observation that non-oxidative conditions favour tar formation (of which phenol is a major component) and that these tar components decrease as oxygen increases (Chen et al. 2008). The wind tunnel and field results tended to cluster together. However, the biplot suggested that the FTIR and GC/FID methods differed, particularly along the H₂ against Other direction, which has been recently corroborated in a separate analysis (Weise et al. 2023).

Fig. 2.

Compositional principal components analysis (PCA) biplot for pyrolysis gases distinguishing observations by location and method (instrument). Note the separation between the non-oxidising (bench) and oxidising (wind tunnel, field) pyrolysis environments. Arrows indicate clr-transformed gas concentrations. Points represent the original samples through their PCA scores. Field = 0.1 ha prescribed burn plots, Ft. Jackson, SC, US.

Geometric mean barplots, showing the average deviations of the individual geometric means relative to the overall means, indicated similar trends (Fig. 3). Deviations by instrument showed that FTIR and GC/FID were similar and opposite the GC/MS means. Note that the deviations labelled GC/MS also include gases measured with GS/TCD. The plot by location showed a similar pattern which also indicated the confounding with the instrument/method. When the wind tunnel and field data were combined as a single ‘oxidising environment’ data set, the deviation plot showed consistent differences with the non-oxidising environment. Examination of the mean fitted values of the five balances (as defined in Table 1) also suggested differences between locations (Table 3). To identify the balances between gases which best separated the oxidising environment from the non-oxidising environment, CDA was performed using the balance values. Fig. 4 shows the resulting first two CDA dimensions (those accounting for the highest fraction of the between-group variability). The CDA scores of the samples at the three locations showed that these locations were essentially arranged along the first CDA dimension (x-axis; 99.9% of the between-group variation accounted for), with this being mostly defined by the contrast between the CO-CO₂ balance (z˜3) and the other balances (except for z˜1 pointing at a nearly orthogonal direction). The mean scores per location, indicated by the large red symbols, also illustrated the separation between the bench scale (with higher values of the z˜3 balance and lower values of the z˜2 ((CO, CO₂) vs. (H₂, CH₄, phenol)) and z˜4 (CH₄ vs (H₂, phenol)) balances) and the other two locations.

Fig. 3.

Geometric mean barplots (in log-ratio scale) by instrument, location and environment for pyrolysis gases measured from wildland fuels. The reference line at zero indicates the overall mean composition.²

Fig. 4.

Plot of first two dimensions resulting from canonical discriminant analysis (CDA) stressing the difference in scientifically meaningful balances (Table 1, z˜k). between the oxidising (wind tunnel, field) and non-oxidising environments (bench). These two dimensions accounted for 100% of original variability. Arrows indicate the balances that separated the two environments. Points indicate CDA scores of the original observations. The large red symbols indicate the mean values of the CDA scores for each group. Field = 0.1 ha prescribed burn plots, Ft. Jackson, SC, US.

PERMANOVA testing the effect of location on the gas composition found that location was highly significant at the usual statistical significance threshold of 0.05 (P-value < 0.001), and pairwise comparison of the individual effects found that the non-oxidising environment results differed from the other two locations (P-values = 0.001), but no significant differences were found between field and wind tunnel (P-value = 0.294). Based on these tests and the evidence provided by the previous exploratory data analyses, the wind tunnel and field data were grouped together as an oxidising environment for further comparison with the non-oxidising environment results.

Fig. 5 summarises boxplots and results from Kruskal–Wallis tests looking at how each balance was affected by the environment. All five balances differed significantly between the oxidising and non-oxidising environments according to the results. For all balances except CO vs CO₂, their values were larger in the oxidising environment. As would be expected, the CO vs CO₂ balance indicated that there was significantly more CO relative to CO₂ in the non-oxidising environment. This has been observed in other less oxidative environments, including the pyrolysis and smouldering stages of fire (Yokelson et al. 1997; Tihay et al. 2009b; El Houssami et al. 2016). The data analyses above all suggested differences between the three locations with the wind tunnel and field locations (oxidising) being more similar compared to the bench (non-oxidising) compositions.

Fig. 5.

Box and whiskers plots for selected balances ( z ˜1… z ˜5) of pyrolysis gases for non-oxidising (fuel rich, no O₂) and oxidising (ambient atmosphere) environments. Adjusted P-values from Kruskal–Wallis testing for comparisons of balances by environment shown. See Table 1 for balance description.

Discussion

A wildland fire is a chaotic event and the gases which are produced by the pyrolysis of wildland fuels may (a) rapidly oxidise producing (and sustaining) the flame, (b) react within the fuel-rich high temperature flame zone to either form lower molecular weight gases or (c) polymerise and condense to form PAH and soot, (d) pass through the flame and only partially oxidise, or (e) cool and condense without passing through the flame zone (Ryan and McMahon 1976) yielding a rich mixture of gases and particulates collectively known as smoke. This last process is prevalent in smouldering combustion. While these chemical processes occur on small time scales as short as a few milliseconds and there are myriad pathways along which the reactions can occur (Sullivan 2017), the data used in the present analysis were collected under largely controlled conditions with the hope that they represented the ‘average’ or ‘most frequent’ composition of gases for these fuels in both non-oxidising and oxidising environments. In the non-oxidising FFB experiments which measured pyrolysis products with a focus on tars and major light gases associated with a range of plant species, different heating rates and modes and foliar moisture contents, differences in composition due to heating mode/rate and species were found (Weise et al. 2022b). Foliar moisture also affected some of the other log-ratios. However, when compared to the oxidising environment compositions, the non-oxidising results were tightly grouped indicating less variability in composition which was not surprising as the oxidising environment was less controlled.

In the oxidising environment experiments, care was taken to capture gases immediately before the flame front arrived, i.e. during the preheating stage. In the wind tunnel experiments, pyrolysis and flaming combustion samples were collected and their compositions were shown to differ (Weise et al. 2022c). The sampling setup was designed to collect flaming combustion and pyrolysis samples at different locations relative to the flame and the convective plume (Aminfar et al. 2020); the sample inlets for the pyrolysis samples were below the convective plume. Relatively more H₂ was observed in the pyrolysis samples. In the oxidising environment, CO₂ was the dominant component (relative amounts of 0.93 and 0.95 for pyrolysis and flaming, respectively). The relatively high CO₂ fraction is consistent with reported CO₂ mass fractions of 0.62–0.72 from pyrolysis studies using the Fire Propagation Apparatus (Tihay et al. 2009b; El Houssami et al. 2016). It is not known if these reported CO₂ mass fractions were based on excess concentrations as is standard practice when handling fire emissions. Prior to flame front arrival, maximum radiometric leaf temperatures adjacent to the FTIR probe in the wind tunnel ranged from 406 to 933 K (Supplementary Table A1 in Weise et al. 2022c) which is well-within the temperature range associated with primary degradation of biomass (Di Blasi 2008). While we are confident that our reported measurements resulted primarily from pyrolysis, we are aware of the idea of buoyant turbulent eddies forcing hot gases in front of the flame (Finney et al. 2015) which could have affected the measured composition by the introduction of flame gases. The background-oriented schlieren imagery used to visualise convective heating of the plants in the wind tunnel experiments did not reveal these forward-thrusting turbulent jets (Aminfar et al. 2020) in the no-wind experiments. This imagery viewed the flow field parallel to the flame front (perpendicular to fire spread direction). While the IR imagery taken vertically looking downward at the fuel bed may suggest the presence of the forward-thrusting turbulent jets (Banach et al. 2021, p. 2368), the camera was looking through the slightly tilted flames which appear to be horizontal jets due to the 2D nature of the imagery. While there is much evidence to suggest that mixing of flame gases did not occur or was minimal during the pyrolysis gas sampling in the wind tunnel experiments, the possibility of this mixing cannot be conclusively eliminated. More detailed measurements, imagery and analysis that include local temperature, velocity, and O₂ measurements in addition to the current measurements might be able to help in determining the physical processes involved in this pre-flame region.

The one thing that seems clear from the data analysis conducted is that pyrolysis products measured in the bench experiments were much richer in H₂ and other hydrocarbons than those in the wind tunnel and field experiments. The small amount of pyrolysis products detected in the wind tunnel and field experiments do not seem to resemble the compositions of similar products in the bench experiments, as indicated by the PCA results. This is likely a result of the instrumentation and methods used in these relatively unique experiments measuring pyrolysis in a wildland flame environment.

Recognising the importance of time in the sequences of chemical reactions that can occur as pyrolysates released into an environment containing O₂ gas, it is difficult to confidently state that the gas samples represent only that stage in the sequences of reactions. This is due in part to the sampling method used for the field experiments (canister samples, static FTIR samples). In the wind tunnel experiments, however, a subset of the FTIR samples were dynamic samples meaning that one could observe the evolution of the gas mixture over time (Banach et al. 2021). As seen in Fig. 6 and explained in (Banach et al. 2021), the dynamic FTIR data were synchronised with thermal imagery of the fuel bed, which was used to identify the different stages of pyrolysis and combustion. Such a temporal series of gas composition has several potential uses. Using CoDA techniques, empirical models of the change in the gas composition could be derived and used for an improved representation of pyrolysis and combustion processes in physical models (e.g. Borujerdi et al. 2022). The temporal series could be compared with the static samples to locate the time step where the dynamic composition is closest to a static sample using some suitable multidimensional distance approach.

Fig. 6.

Burn 87 from project RC2640, inkberry with longleaf pine needles: (a) CO and CO₂ spectral profile from 2250 to 2000 cm⁻¹. Purple traces indicate the ambient stage, green and orange traces indicate the pre-combustion/pyrolysis stage, red spectral traces indicate the flaming stage and blue traces indicate smouldering. (b) Largely C₂H₄ (ethylene)spectral waterfall plot from 1000 to 900 cm⁻¹ with accompanying C₂H₄ reference spectrum is the black trace. Spectra are recorded/produced every 0.79 s (Banach et al. 2021).

In physical models, the gas composition from a non-oxidising environment can be the starting point to represent the pyrolysis gas mixture and to predict the evolution of the gas mixture to the point of ignition. However, possible modifications to the modelling may be needed due to oxidative pyrolysis. The gas composition from the oxidising environment could serve as validation data for a physical model describing pyrolysis in an oxidising environment. As the composition evolves in time, identifying the time step in the simulation where the multidimensional distance between the modelled composition and the measured composition is minimised for a static FTIR sample (Banach et al. 2021) could prove to be informative. If a dynamic gas composition was available, such a comparison would be between the modelled and observed time sequences of gas composition. These are just a few possibilities of the future work that needs to be performed to support the continued use of non-oxidising environment pyrolysis data as the starting point for physical modelling of fire.

Conclusions

Data from a series of carefully controlled experiments measuring the composition of pyrolysis gases in non-oxidising and oxidising environments, determined using gas chromatography and FTIR spectroscopy that have been previously reported, were analysed within a compositional data analysis framework. Principal component analysis, canonical discriminant analysis and permutational analysis of variance showed that the subcomposition of gases (H₂, CO, CO₂, CH₄, phenol, and other gases) common to all three experiments differed significantly between the non-oxidising flat-flame burner environment and the oxidising environments of small fires in a wind tunnel and in prescribed burns at Ft. Jackson, SC, US. Results seem to indicate the need for more fundamental research on the early time-dependent pyrolysis of vegetation in the presence of oxygen in an environment where a flame is the primary heat source rather than using a bulk composition for the entire pyrolysis process taken from non-oxidising experiments. The complete composition of measured gases resulting from pyrolysis of live foliage from common shrub and tree species in the southeastern United States is available to be used in physically based models of fire to model pyrolysis.