Pyrogenic organic matter production from wildfires: a missing sink in the global carbon cycle.

Wildfires release substantial quantities of carbon (C) into the atmosphere but they also convert part of the burnt biomass into pyrogenic organic matter (PyOM). This is richer in C and, overall, more resistant to environmental degradation than the original biomass, and, therefore, PyOM production is an efficient mechanism for C sequestration. The magnitude of this C sink, however, remains poorly quantified, and current production estimates, which suggest that ~1-5% of the C affected by fire is converted to PyOM, are based on incomplete inventories. Here, we quantify, for the first time, the complete range of PyOM components found in-situ immediately after a typical boreal forest fire. We utilized an experimental high-intensity crown fire in a jack pine forest (Pinus banksiana) and carried out a detailed pre- and postfire inventory and quantification of all fuel components, and the PyOM (i.e., all visually charred, blackened materials) produced in each of them. Our results show that, overall, 27.6% of the C affected by fire was retained in PyOM (4.8 ± 0.8 t C ha(-1)), rather than emitted to the atmosphere (12.6 ± 4.5 t C ha(-1)). The conversion rates varied substantially between fuel components. For down wood and bark, over half of the C affected was converted to PyOM, whereas for forest floor it was only one quarter, and less than a tenth for needles. If the overall conversion rate found here were applicable to boreal wildfire in general, it would translate into a PyOM production of ~100 Tg C yr(-1) by wildfire in the global boreal regions, more than five times the amount estimated previously. Our findings suggest that PyOM production from boreal wildfires, and potentially also from other fire-prone ecosystems, may have been underestimated and that its quantitative importance as a C sink warrants its inclusion in the global C budget estimates.

Introduction

Wildfires burn, on average, 464 Mha (∽4%) of the Earth*s vegetated land surface every year and emit 2.5 Pg of carbon (C) to the atmosphere (Randerson et al., 2012), which is equivalent to a third of the current emissions from fossil fuel consumption (Boden et al., 2012). The total area burned and fire intensity are projected to increase in a warming climate, which in turn will increase C emissions from wildfires (IPCC, 2013). Notwithstanding this, on the timescale of decades to centuries, wildfires are considered ‘net zero C emission events’, because the C emitted is balanced by C uptake by regenerating vegetation (excluding deforestation and peatland fires; Bowman et al., 2009; IPCC, 2013). This ‘zero C emission’ scenario, however, is likely to be flawed, as it does not consider the production of pyrogenic organic matter (PyOM; also known as charcoal, ‘pyrogenic carbon’ or ‘black carbon’). During each wildfire, instead of being emitted to the atmosphere as CO2, other gases and as aerosols, a fraction of the burning biomass is converted to PyOM, which is a continuum from partly charred organic materials through charcoal to soot (Bird & Ascough, 2012).

PyOM is C-enriched and has an enhanced resistance to degradation, with residence times in the environment generally one or two orders of magnitude longer than its unburnt precursors (Schmidt et al., 2011; Singh et al., 2014). Therefore, PyOM production during wildfire could function as a long-term (decades-millennia) C sequestration mechanism (Lehmann et al., 2008; Reichstein et al., 2013; Ottmar, 2014). In this context, the application of ‘man-made PyOM’ (biochar) to soils is currently seen as one of the most viable global approaches of offsetting C emissions (Woolf et al., 2010). However, despite its acknowledged importance as a C sink, the role of wildfire PyOM in the global C balance remains contentious because of uncertainties in its production and fate (Forbes et al., 2006; Preston & Schmidt, 2006). The inclusion of PyOM in C estimates may be crucial for accurate projections of future climate change (Lehmann et al., 2008) and might explain part of the elusive ‘missing C sink’ of ∽1.5 Pg C yr−1 (Lal, 2008). To enable the inclusion of PyOM in C accounting, accurate quantification of the conversion of C from the fire-affected fuel (CA) to PyOM (PyC) during wildfires is required. The necessary complete quantification of the PyOM produced by wildfire, and the C retained in it (pyrogenic C, i.e., PyC), however, has never been achieved.

Previous research has pointed to a very limited proportion of the C affected by fire being converted to PyOM (∽1–5%, see reviews by Preston & Schmidt, 2006 and Forbes et al., 2006). The data underpinning this notion, however, are rather limited and not always representative. For example, much of the data obtained to date have been based either on laboratory experiments (e.g., Brewer et al., 2013) or prescribed fires (e.g., Alexis et al., 2007), which are ignited under controlled conditions for management purposes and are usually not representative of wildfire conditions (Urbanski, 2014), or on wildfires, which, because of their unpredictability, do not allow the prefire sampling of fuels (i.e., biomass, necromass and soil organic matter) necessary for calculating budgets (Forbes et al., 2006). Other related research has focused on centennial-scale in-situ accumulation of PyOM in soils (e.g., Ohlson et al., 2009; Kane et al., 2010), which does not account for the fact that much of the PyOM produced is often rapidly removed from burnt sites by wind and water erosion (Santín et al., 2012; Bodí et al., 2014). Furthermore, some studies have only focused on specific chemically defined PyOM fractions rather than on the complete range of PyOM components (e.g., Kuhlbusch et al., 1996; Czimczik et al., 2003). Others have quantified PyOM in only some of the fuel components where it is produced (e.g., down wood, Donato et al., 2009; aboveground fuels, Worrall et al., 2013).

To achieve a complete quantification of wildfire PyOM production with respect to fuel affected by wildfire, we utilized an experimental forest fire that represented wildfire conditions. This allowed, for the first time, quantification of (i) all fuel components, and their respective C contents, before and after fire, and (ii) all the PyOM components, and their respective C contents, found in-situ immediately after fire.

Materials and methods

Study site and experimental forest fire

The Canadian Boreal Community FireSmart Project site is located 40 km north of Fort Providence (61°34′55′’ N; 117°11′55′’ W; Northwest Territories, Canada) and was also the location of the International Crown Fire Modelling Experiment (1995–2001; Stocks et al., 2004). Stand-replacing fires, aimed to be representative of wildfires, are carried out here by FPInnovations Wildfire Operations Research in close collaboration with the NT Government to address wildfire management issues.

The experimental plot selected to be burnt was a 1.7 ha mature stand of jack pine (Pinus banksiana) with a tree age of 80 years, a tree density of approximately 7600 stems ha−1 and average tree height of 14 m. The fire was started at 16:00 hours on 23 June 2012. The ambient temperature was 28 °C and relative humidity was 22% with winds of 10–12 km h−1. The last rain had occurred 6 days prior (0.5 mm), with a total precipitation over the previous month of 4.3 mm. At the time of the fire, the component values of the Canadian Forest Fire Weather Index System (Van Wagner, 1987) were: Fine Fuel Moisture Code 92.9; Duff Moisture Code 135; Drought Code 394; Initial Spread Index 9.5; Buildup Index 145 and Fire Weather Index 35. The experimental burn resulted in a high-intensity crown fire with a head fire intensity of 8000 kW m−1, a flame height of 5-6 m above canopy level and a spread rate of ∽6–7 m min−1 (Fig.1). This fire behaviour is in the typical range for boreal crown fires (de Groot et al., 2009), with the head fire intensity matching the average estimated for June in the boreal forests of Western Canada for 2001–2007 (de Groot et al., 2013).

Figure 1

FireSmart experimental forest fire (June 2012). This stand-replacing high-intensity crown fire (head fire intensity 8000 kW m−1) reproduced typical boreal wildfire conditions (de Groot et al., 2009, 2013).

Experimental design and sampling

Before the fire, three parallel transects of 20 m length with nine sampling points each (every 2m; n = 27) were established in the centre of the plot in the direction of the prevailing wind (E–W). All sampling points were instrumented with thermocouples and auto-loggers (Lascar, Easylog, Wiltshire, UK) to record temperatures at the forest floor surface and at the interface between forest floor and the mineral soil (Santín et al., 2013).

A detailed prefire inventory and sampling of all fuel components: forest floor, down wood (i.e., woody debris on the ground), overstory, understory and mineral soil was carried out. Immediately after the fire (within 1–6 days), inventorying and resampling of each fuel component were repeated distinguishing between (i) remaining uncharred fuel and (ii) PyOM produced (see methodological details for each component in the following section). PyOM was visually identified, and sampled, as all blackened materials, i.e., charred, and, thus, chemically altered by fire (Bird & Ascough, 2012). This visual identification allows accounting for the whole range of the PyOM materials remaining in-situ immediately after the fire. Pyrogenic aerosols emitted within the smoke, and thus immediately transported ex-situ, are not examined in this study.

All fuel and PyOM samples were oven-dried (65 °C) and cleaned by hand to remove contamination, such as soil particles in the forest floor samples or unburnt material within the PyOM components. The dry weight of all samples was then recorded and subsamples were ground for C quantification. Total C content was determined in duplicate by quantitative high temperature combustion and conversion to CO2 using an ANCA GSL elemental analyser interfaced with a Sercon 20/20 mass spectrometer. Presence of carbonates was tested by addition of 10% HCl to a set of representative PyOM subsamples. No effervescence was observed, what indicates a very low (<1%) concentration of carbonates (Rayment & Lyons, 2011). Therefore, the inorganic C concentration in the studied samples is considered negligible and total C as being equivalent to total organic C.

For each fuel component, mass and C loads (t ha−1) in prefire fuel and postfire uncharred fuel and PyOM were calculated from field measurements, dry weights and measured C concentrations (see methodological details in the following subsection). To provide a measure of the uncertainty of our estimations (Table1), the confidence intervals (CIs) of the estimations were derived by applying studentized bootstrap procedures for the mean (Efron, 1982). This statistical approach was chosen instead of the classical Gaussian approaches due to the asymmetry of the variables analysed and the size of the datasets, which were not large enough to balance the lack of symmetry. The studentized bootstrap CIs were thus implemented for a linear combination of independently sampled variables using the software R (version 3.1.1). All the CIs were determined at a confidence level of 95% and by using B = 10000 bootstrap iterations. Mass and C losses were calculated, for each fuel component, as the difference between prefire fuel and the remaining uncharred postfire fuel + PyOM (i.e., Fuel lost = Prefire fuel–Postfire uncharred fuel–PyOM). The conversion rate of C in fire-affected fuel (CA) to C in PyOM (PyC) was subsequently estimated (i.e., PyC/CA), with CA being the sum of PyC + C lost.

Table 1

Mass and C loads in fuel and pyrogenic organic matter (PyOM) components before and after fire (± bootstrap confidence intervals at 95%). PyC/CA is the conversion rate of C in fire-affected fuel (CA) to C in PyOM (PyC), with CA being the sum of PyC + C lost

Component			Postfire
	Prefire		Uncharred		PyOM		Lost*
	Mass (t ha−1)	C (t ha−1)	Mass (t ha−1)	C (t ha−1)	Mass (t ha−1)	PyC (t ha−1)	Mass (t ha−1)	C (t ha−1)	PyC/CA (%)
Forest floor	45.2 ± 10.3	19.7 ± 6.2	26.3 ± 3.9	9.9 ± 1.7	3.6 ± 0.6	1.9 ± 0.4	12.6 ± 8.3	6.0 ± 4.4		24.5
Down wood	38.8 ± 13.0	17.9 ± 6.0	32.7 ± 11.7	15.1 ± 5.4	1.9 ± 0.2	1.4 ± 0.2	4.0 ± 1.1	1.3 ± 0.5		51.2
Overstory
(i) Bark	7.2 ± 3.4	3.4 ± 1.5	2.6 ± 2.1	1.2 ± 1.0	2.5 ± 1.3	1.5 ± 0.8	2.5 ± 3.2	0.8 ± 1.6	67.1
(ii) Needles	10.1 ± 1.5	5.4 ± 0.8	0.0 ± 0.0	0.0 ± 0.0	0.6 ± 0.0	0.4 ± 0.0	9.5 ± 1.5	5.1 ± 0.8	7.3
Total†	95.7 ± 15.2	42.8 ± 7.3	59.9 ± 12.2	25.3 ± 5.6	7.9 ± 1.3	4.8 ± 0.8	27.6 ± 8.7	12.6 ± 4.5	27.6

Lost refers to what has been emitted to the atmosphere or, in the case of individual components, may include some transfer between components.

Note that the total values have been calculated by applying studentized bootstrap procedures for the means.

Quantification of prefire and postfire fuel and PyOM components

Detailed methodologies for quantification and sampling of the main fuel and PyOM components, pre- and postfire, are described in the following subsections.

Forest floor

Before the fire, bulk forest floor samples were taken using 20 × 20 cm sampling squares along two parallel lines between the three sampling transects (n = 10). The total depth of the forest floor was measured at each corner of the square and the entire layer was carefully collected. The forest floor was mainly composed of litter, mosses, lichens, needles, duff and humidified organic material. Cones and all woody debris <0.5 cm diameter were also considered as part of the forest floor.

After the fire, the charred forest floor layer (i.e., ash layer) and the uncharred layer underneath were sampled along the three sampling transects at every sampling point (i.e., every 2 m; n = 27). This ‘charred layer’ or ‘ash layer’ comprises both organic (i.e., PyOM) and mineral residues resulting from the burning of forest floor and aboveground inputs (Bodí et al., 2014). Wildfire ash colour can vary widely, from light to dark, depending on the fuel affected and formation conditions (Bodí et al., 2014), but all ash, irrespective of its colour, is of pyrogenic origin. Therefore, in our study, all ash (even when not black) was sampled and included in our PyOM inventory.

The charred forest floor layer was sampled using a 30 × 30 cm square. The depth of the charred layer was measured at each corner of the square and the entire layer was carefully collected. For the uncharred layer beneath the charred layer, the same procedure was followed in a subsquare of 10 × 10 cm. The samples of charred forest floor were manually cleaned from any visually uncharred materials (<7% dry weight) derived from the uncharred forest floor layer underneath. The uncharred forest floor samples were cleaned of any charred particles and mineral soil (<6% dry weight). Following drying and cleaning, all samples were weighed and subsamples ground for C analyses.

Prefire mass and C loads and postfire uncharred and PyOM mass and C loads were calculated for the individual samples using their density and C contents values. Afterwards, the pre- and postfire forest floor loads (Table1) were estimated by applying a studentized bootstrap CI for the mean on the obtained transformed samples. For calculating the lost mass and C loads (Table1), a studentized bootstrap CI for the difference of means was used (i.e., prefire vs. postfire (uncharred + PyOM) components).

During the fire, particle traps were used to capture (and allow discounting of) any aboveground contribution to the charred layer present on the forest floor (procedure modified from Lynch et al., 2004): before the fire, 17 aluminium trays (333 cm2 each) filled with water were placed flush with the forest floor along two parallel transects at the Southern (nine trays) and Northern (eight trays) ends of the sampling transects, orientated in the direction of fire propagation. After the fire, the contents of all trays were collected by sieving (>0.2 mm) and combined to generate a composite sample. Following drying (65 °C), any uncharred material (e.g., brown needles and/or uncharred twigs and cones) was removed from this composite sample and, afterwards, its C content determined. The PyOM mass per unit area of material collected in the traps was deducted from the total production of PyOM calculated for the forest floor. The total PyOM quantified in the particle traps was assigned to the ‘Overstory’ component (see subsection below). This reduction in PyOM quantity for the forest floor may be somewhat too high as some of the material in the trays may have been derived from the forest floor and lifted into the particle traps by convection currents during the fire. This may result in an underestimation of the PyOM values for the forest floor. A postfire contribution to the forest floor not captured in the traps were detached small PyOM particles from charred down wood (i.e.,charcoal fragments ≤0.5 cm in size and therefore sampled as part of the postfire forest floor). It is not possible to give a reliable estimate for this potential addition. Nevertheless, the associated potential overestimation of PyOM in the forest floor component would not affect the total PyOM production estimate from this fire, as these small PyOM particles are accounted for in the forest floor, instead of the down wood component.

Down wood

Before the fire, the line intersect method (LIM) was employed to calculate loads (t ha−1) of dead wood (twigs, limbs, branches and logs) on the ground (Alexander et al., 2004). This involved counting the number of down wood pieces intercepting the three 20 m sampling transects (i.e. total length 60 m) using the following roundwood diameter size classes: 0.5–1.0, 1.1–3.0, 3.1–5.0 and 5.1–7.0 cm. These are referred to as Classes II-V (note that Class I, ≤0.5 diameter, is not considered in the down wood component as it was sampled and included within the forest floor). For downed logs >7.0 cm in diameter, the diameters of all pieces intersecting the transects were measured and recorded as either sound or rotten according to the degree of decay. When applying the LIM, the following standard principles were followed: trees are considered down if they lean >45°; branches still attached to standing trees are not considered; curved twigs which intersect >1 times the transect are counted at each intersection; pieces that fall perfectly in line with the transects are not tallied (Alexander et al., 2004).

To calculate down wood loads Eqn (1) was used for size classes II–V:

where W = down wood loads (t ha−1), G = specific gravity (g cm−3), h = piece tilt angle (degrees), n = number of intercepts over the length of the transects, QMD = quadratic mean diameter (cm), and L = length of transects (total length 60 m). The values used for G, h and QMD are those given in Nalder et al. (1999). For roundwood pieces >7.0 cm in diameter, Eqn (2) was used to calculate loads:

where W = down wood loads (t ha−1), Σd2 = sum of the squared diameters for intercept pieces (cm2), G = specific gravity (g cm−3) with different values for sound and rotten pieces according to Delisle & Woodard (1988), and L = length of transects (total length 60 m). The moisture content of rotten pieces >7.0 cm before the fire was too high for ignition. They were therefore excluded from postfire calculations.

According to Eqns (1) and (2), the contribution to W of each down wood piece intercepting the sampling transects can be computed in such a way that W is the sum of the individual contributions. Using these individual contributions, a studentized bootstrap for the total W (with the mean adjusted by the sample size) was applied to estimate total prefire mass loads in Table1. Subsequently, to estimate total prefire C loads (Table1), the mass of each individual contributions (i.e., pieces) was multiplied by the C content of representative down wood samples (Table2), and, in the same way as for mass loads, a studentized bootstrap for the total was applied.

Table 2

Average C concentrations in prefire fuels and postfire uncharred fuels and PyOM. Values are given as the arithmetic mean ± standard error of the mean, number of samples is given in brackets

Component	Prefire	Postfire
		Uncharred	PyOM
	C (g g sample−1)	C (g g sample−1)	C (g g sample−1)
Forest floor	0.405 ± 0.020 (10)	0.369 ± 0.020 (27)	0.541 ± 0.020 (27)
Down wood	0.462 ± 0.000 (3)	*0.462 ± 0.000 (3)	0.729 ± 0.038 (3)
Overstory
(i) Bark	0.473 ± 0.017 (10)	*0.473 ± 0.017 (10)	0.629 ± 0.012 (10)
(ii) Needles	0.544 ± 0.010 (8)	n.a.	0.680 ± 0.002 (6)

Prefire values are used; n.a.: not applicable.

After the fire, the LIM was repeated. The accurate relocation of the transects was facilitated by metal pins placed prior to the fire. During postfire sampling, we observed an increase in the number of down wood pieces for some size classes compared to prefire values (e.g., for size class IV a total of 18 pieces were detected along the three transects before the fire and 29 after the fire). This is likely due to removal of the upper part of the forest floor by fire exposing previously embedded, and hence not recorded, down wood (Volkova & Weston, 2013). These ‘buried pieces’ are not accounted for in down wood inventories (Brown, 1974). To overcome this postfire shortcoming of the LIM method, we used the following alternative sampling approach for determining the amounts of PyOM produced from the down wood component: a representative initial area of 40 m2 was selected 20 m east of the experimental transects, and all down wood pieces present after the fire were examined. For each piece, its diameter was recorded and the depth of charring measured, both on the side with the deepest charring depth and the opposite side. Pieces were either broken or sawn in half to measure charring depth. For size classes IV and above, the sampling area examined was increased to reach a minimum of 60 pieces for each size class. From these data, an average charring depth was assigned to each size class (note that size class I was not examined as it was sampled and included within the forest floor component).

After the fire there was no evidence of log trenches or shallows, which are indicators of complete combustion (Tinker & Knight, 2000). Hence it was reasonable to assume no complete combustion occurred for the larger classes (>3 cm diameter). Complete or nearly-complete combustion of smaller size classes may have occurred, but any small PyOM particles (≤0.5 cm diameter) remaining on site from these would have been sampled as part of the forest floor. No standing trees fell during or immediately after the fire, so contribution from standing trees to down wood was considered to be zero.

The postfire PyOM loads in down wood (Table1) were estimated as for the prefire mass loads (i.e., studentized bootstrap for the total), but taking into account the measured mean charring depth for each size class (instead of total diameters) as well as a G value of 0.23 g cm−3 (±0.02 SEM-standard error of the mean) obtained from representative charred down wood samples collected in the field (n = 13). C loads in PyOM (Table1) were then calculated by multiplying PyOM mass loads in each class size by the measured C content of representative charred down wood samples (Table2) and computing again a studentized bootstrap CI estimation. For estimates of the proportion of down wood that had been lost during combustion (Table1) we used the average proportions of 8.8% for coarse down wood (>7 cm diameter) and 16.1% for fine down wood (<7 cm diameter), with respect to prefire loads, measured during an experimental fire of similar characteristics (head fire intensity ∽8000 kW m−1) in a mature jack pine stand (Stocks, 1989) and the corresponding studentized bootstrap CIs. Subsequently, total uncharred down wood remaining after the fire was estimated by applying a studentized bootstrap CI to the difference between prefire down wood and down wood converted to PyOM + down wood lost (note that all of these measurements are performed on the same individuals). C loads in the postfire uncharred downwood were calculated in the same way as the prefire C loads, i.e., by multiplying each uncharred mass individual contribution by the C content of representative uncharred (prefire) downwood samples (Table2), and by applying a studentized bootstrap for the total (Table1). Finally, the C lost was calculated by computing a studentized bootstrap mean CI to the difference between C in the prefire down wood and C remaining in postfire uncharred down wood + C in PyOM (note that also these calculations are applied on the same individuals).

Overstory

To calculate tree density, overstory trees (i.e., trees with a Diameter at Breast Height measured Outside the Bark (DBHOB) >3.0 cm) were inventoried using the point-centred quarter method (Alexander et al., 2004). This involved measuring the distance from each sampling point (i.e., every 2 m along the three sampling transects) to the nearest tree in each of the four quarters of an imaginary square, the centre of which is the sampling point. In addition, for each tree, the species and condition (live or dead) were recorded and the DBHOB measured. If in a given quarter there was no tree <5 m from the centre, that quarter was recorded as ‘no tree’, and, when calculating tree density, established correction factors were applied (Mitchell, 2007).

Tree density (0.76 stems m−2) was calculated following Eqn (3):

where D = overstory tree density (number of stems m−2), and AD is the average distance (m) from the closest tree to the sampling point (i.e., 1.60 m ± 0.09 SEM; n = 108).

All trees were killed by the fire but none fell during or immediately after the fire so total tree density pre- and postfire was considered to be the same.

Overstory I: stems

Standing tree stems suffered almost exclusively only charring of the bark. Evidence of wood charring was only detected in some pre-existing snags. Therefore, we focused on bark as the main component for PyOM production from stems. Before the fire, bark was sampled from representative trees (n = 10) outside the sampling area so that bark sampling did not affect fire behaviour within the burnt plot. For each tree, DBHOB and height was recorded and the entire bark layer was scraped at breast height from an area of 20 cm height around the whole trunk (i.e., a 20 cm strip of bark taken at +/− 10 cm from DBHOB height). After the fire, bark was sampled at representative trees (n = 10) within the three experimental transects using the same procedure as before the fire, but with separating the charred and the uncharred layers of bark. The sampled trees were subsequently felled to facilitate recording of (i) total tree charring height and (ii) fire effects on the canopy (i.e., needles and branches). Total tree height and DBHOB were also recorded. Bark samples (prefire and postfire charred and uncharred) were oven dried (65 °C), weighed and a subsample ground for C analyses.

Total prefire bark mass per tree was estimated by multiplying the measured mass of bark (g m−2) by the total tree surface area (m2) calculated by assuming a conical tree shape. The estimated total bark mass per tree was multiplied by the average total tree density in the experimental plot (0.76 stems m−2) to obtain an estimation of prefire bark loads (t ha−1) based on each sampled tree. The final bark load estimation (Table1) was obtained by applying a studentized bootstrap CI mean estimation on the obtained transformed sample. Similar calculations for postfire sampled trees were done, distinguishing between the charred and uncharred bark mass per tree by using the total height as well as the height of charring. In our approach, we assume homogeneous charring of the bark along the whole height of charring, although it is conceivable that the charring degree of the bark varies along the trunk, with deeper charring at the base of the tree, where the bark is in contact with ground fuels.

To obtain prefire bark C loads, the C content in bark measured for each prefire tree was multiplied by the corresponding total dry prefire bark load estimation. For postfire charred bark (PyOM) the same procedure was applied by using the C content measured for each postfire sample, and distinguishing between charred and uncharred bark (for uncharred bark the average C content value for prefire bark was used, see Table2). As before, the final total estimation (Table1) was obtained by applying a studentized bootstrap CI mean estimation on each obtained transformed C sample. Finally, ‘lost bark mass’ (Table1) was computed by applying a studentized bootstrap CI for the difference of means for independent samples [i.e., prefire mass vs. postfire (charred + uncharred) mass]. An analogous procedure was applied for estimating ‘lost bark C’ [prefire C vs. postfire (charred + uncharred) C].

Overstory II: canopy

For the type of forest fire investigated here, crown fuels are commonly considered to be limited to needles and dead branchwood material <1 cm diameter (Stocks, 1989). Visual examination of the felled trees showed that needles were the principal crown component burnt whereas branches and twigs suffered minimal charring. Therefore, only needle loads were considered when accounting for PyOM production from the canopy. Prefire needle fuel loads were derived using the regression Eqn (4) developed for this experimental site during the International Crown Fire Modelling Experiment (Alexander et al., 2004):

where Y is total dry needle weight (kg) and X is DBHOB.

Needle fuel loads were calculated according to this equation for each (live) tree for which DBHOB had been measured for the calculations of ‘overstory tree density’ described at the beginning of the ‘Overstory’ subsection (n = 38). Needle weights were then converted to loads (t ha−1) using the (live) tree density in the experimental plot (0.40 stems m−2) and a studentized bootstrap CI for the mean was applied (Table1). C loads were then calculated by multiplying the total needle loads by the C content determined from fresh needle samples (Table2) and, as before, a studentized bootstrap CI for the mean was applied (Table1). After the fire, only a few needles remained in the trees, which were dead, but not charred (i.e., brown due to heat, but not chemically altered by the fire). For calculations (Table1), we therefore assumed that essentially no needles remained in the crown after the fire (either charred or uncharred). Thus, any PyOM produced from needles either fell on the ground or was lost from the system, i.e., transported ex-situ within smoke during fire. To estimate the PyOM production from this component, the PyOM quantified in the particle traps placed on the forest floor was entirely attributed to canopy contribution (see ‘Forest floor’ subsection above). It is conceivable that the material captured in the trays also included a contribution from charred bark from standing stems and, as stated before, also particles lifted from the forest floor by convection currents during the fire. Thus, the in-situ PyOM deposition from needles estimated here is probably an overestimate of the aboveground contribution, but will not affect the total PyOM production estimate.

Understory

Before the fire, the understory inventory involved recording the diameters at breast and ground height (cm) and the height (cm) of any understory stems (i.e., saplings and shrubs <3.0 cm DBHOB) at every sampling point along each transect using a 1 m radius fixed plot (Alexander et al., 2004). Understory vegetation, however, was so scarce (<0.1 stems m−2) that this component was considered irrelevant and not further included in this study.

Mineral soil

Before the fire, bulk samples of the mineral soil were taken using a 5 × 5 cm soil corer (n = 10) along two parallel lines between the three sampling transects, at the same sampling points as for the prefire forest floor sampling. The mineral soil sampled was a waterlogged, stony sandy loam derived from fluvio-glacial deposits. The fire did not affect the mineral soil (maximum temperature recorded at mineral soil surface was <70 °C, n = 27). Therefore, soil was not a relevant fuel component for this study and not further considered.

Results

Forest floor

The forest floor had an average depth to the mineral soil of 6.5 cm (±0.3 SEM; n = 108) and is the studied fuel component storing the largest amount of C before the fire (19.7 ± 6.2 t ha−1, Table1). During the fire, the mean maximum temperature at the surface of the forest floor was 750 °C (range 550–976 °C, n = 27) with a mean residence time >300 °C of 180 s (range 65–365 s) (see Santín et al., 2013 for further details). This resulted in an average charring depth of the forest floor of 3.9 cm (±0.2 SEM; n = 108) and generated a continuous layer of charred material, i.e. ash (1.3 cm avg. depth ± 0.1 SEM; n = 108) (Fig.2). This charred layer is C enriched (0.541 ± 0.020 g g−1) compared both to the prefire forest floor (0.405 ± 0.020 g g−1) and the postfire uncharred layer (0.369 ± 0.020 g g−1; Table2). Overall, 1.9 ± 0.4 t C ha−1 were converted to PyOM within the forest floor, whereas 6.0 ± 4.4 t ha−1 were lost (i.e., potentially emitted to the atmosphere) (Table1). This translates into a conversion rate of 24.5% PyC/CA for the forest floor. It is important to note, as is common for boreal forest wildfires (de Groot et al., 2009), that only the upper part of the forest floor was affected by fire; i.e., only 42% of the total prefire fuel load (19.7 ± 6.2 t ha−1, Table1).

Figure 2

The FireSmart experimental forest fire enabled prefire (a) and immediate postfire (b) inventory and sampling of fuel components (i–v in a), and of their respective amounts of pyrogenic organic matter produced (i–v in b). Pyrogenic organic matter production is given for each of the fuel components in t C ha−1, and also as the ratio of C converted to pyrogenic organic matter with respect to C affected by fire [%] (see Table1 for more details). n.r.: not relevant in this fire.

Down wood

Prefire down wood loads were high (38.8 ± 13.0 t ha−1, Table1), with the greatest contribution being from the bigger size classes: 0.69 t ha−1 (class II), 1.38 t ha−1 (class III), 2.66 t ha−1 (class IV), 4.01 t ha−1 (class V), 27.30 t ha−1 (>7 cm, sound) and 2.91 t ha−1 (>7 cm, rotten). However, the smallest down wood pieces were those most affected by fire, showing average charring depths of 2.60 mm (±0.05 SEM, n = 122) for size class II and 8.10 mm (±0.03 SEM; n = 85) for size class III. None of the thicker pieces (classes IV and above) were completely carbonized, with average charred depths of 1.86 mm (±0.20 SEM; n = 60) and 1.89 mm (±0.25 SEM; n = 60) for classes IV and V, respectively. For coarse woody debris (i.e. >7 cm diameter, sound), the charring was even shallower [0.86 mm (±0.13 SEM; n = 60)].

Overall, 32.7 ± 11.7 t ha−1 of the down wood mass remained unaffected by the fire, 1.9 ± 0.2 t ha−1 of the down wood was converted into PyOM and 4.0 ± 0.2 t ha−1 was lost (Table1). The PyC/CA rate for the down wood component was 51.2% (Table1, Figure2), twice that obtained for the forest floor. Down wood PyOM showed the greatest C enrichment of all fuel components compared to the unburnt precursors, from 0.462 ± 0.00 to 0.729 ± 0.038 g g−1 (Table2).

Overstory

The total prefire loads estimated for the overstory component are not equivalent to total tree biomass because they do not include the standing timber, which was not affected by fire. Prefire loads for needles (10.1 ± 1.5 t ha−1) were higher than for bark (7.2 ± 3.4 t ha−1) (Table1), however, the PyOM production was much higher for bark than for needles, both in terms of total amounts of C (1.5 ± 0.8 and 0.4 ± 0.0 t PyC ha−, respectively) and as PyC/CA rate (67.1% and 7.3%, respectively) (Table1, Fig.2). C enrichment in PyOM for needles involved an increase from 0.544 ± 0.01 to 0.680 ± 0.002 g g−1 and for bark from 0.473 ± 0.017 to 0.629 ± 0.012 g g−1 (Table2).

Discussion

This typical boreal forest fire produced 7.9 ± 1.3 t ha−1 of PyOM (containing 4.8 ± 0.8 t PyC ha−1) and converted 27.6% of the CA to PyC (Table1). The three main fuel components, i.e., forest floor, down wood and overstory (bark and needles combined), produced broadly similar amounts of PyC (1.9 ± 0.4, 1.4 ± 0.2 and 1.9 ± 0.8 t ha−1 respectively; Table1). The PyC/CA conversion rates, however, show substantial differences. For down wood and bark, over half of the CA was converted to PyC whereas for forest floor, only a quarter of the CA was converted to PyC. For needles only 7% of the CA was converted to PyC, with the rest having been combusted, which is broadly in agreement with the assumptions of complete combustion of previous studies (Knorr et al., 2012). Needles and forest floor materials have a much higher surface area to volume ratio than tree stems (bark) and down wood, facilitating access to oxygen and hence greater combustion completeness, resulting in lower PyC/CA conversion for the fuel components with relatively high surface areas (Kuhlbusch & Crutzen, 1995). The lower combustion completeness of fuel with a low surface area to volume ratio is also reflected in the greater relative C enrichment of the PyOM produced within these fuel components (Table2). This lower combustion completeness could be also related to the different chemical composition of the fuel components, with materials richer in lignin such as pine wood (Räisänen & Athanassiadis, 2013) suffering relatively smaller mass losses during charring (Czimczik et al., 2002; Cornwell et al., 2009).

The overall PyC/CA conversion rate of 27.6% found here is substantially higher than previous estimates for boreal regions and other ecosystems elsewhere (∽1–5% PyC/CA; reviews by Kuhlbusch & Crutzen, 1996; Forbes et al., 2006; Preston & Schmidt, 2006). Table3 summarizes the outcomes of 31 studies quantifying PyOM production in different ecosystems and using various approaches. We suggest three main reasons for the substantially lower estimations in these previous studies.

Table 3

Previous estimates of pyrogenic organic matter (PyOM) production from wildfires, prescribed and experimental fires. PyOM production is given as the ratio of C converted to PyOM (PyC) with respect to C affected by fire (CA) [% PyC/CA], and as the quantity of PyC produced (t PyC ha−1). Not included are studies (i) focusing on long-term PyOM pools, (ii) summarizing previous work on PyOM production and (iii) those dealing with atmospheric black carbon or biochar

Study	Ecosystem	Type of fire	PyOM production [%PyC/CA (t PyC ha−1)]	PyOM component(s) studied	PyOM detection and quantification††
Clark et al., 1998	Boreal‡‡	Experimental (wildfire)	2 (0.7)*	Airbone particles	Visual & gravimetric
Czimczik et al., 2003	Boreal‡‡	Wildfire	0.7 (0.06)†	Forest floor	Molecular markers (Benzenopolycarboxylic acids)
Lynch et al., 2004	Boreal‡‡	Experimental (wildfire)	2 (0.58)*	Airbone particles	Visual & gravimetric
Makoto et al., 2012	Boreal‡‡	Wildfire	n.d. (0.25–0.06)*	Bark on standing snags	Visual & gravimetric
Ohlson & Tryterud, 2000	Boreal‡‡	Experimental (wildfire)	n.d. (0.23)*	Airbone particles	Visual & gravimetric
Pitkänen et al., 1999	Boreal§§	Prescribed (slash-and-burn)	n.d. (1.4)*	Airbone particles	Visual & gravimetric
Brewer et al., 2013	Temperate‡‡	Laboratory (fuel beds)	7.2–8.7 (2.0–2.5)	All fuels selected	Visual, gravimetric & total C quantification
Finkral et al., 2012	Temperate‡‡	Prescribed (slash pile)	1–5 (0.05–0.21)	All	Visual, gravimetric & total C quantification
Pingree et al., 2012	Temperate‡‡	Wildfire & prescribed	‡1–8 (n.d.)	Soil§	Chemical (peroxide-acid digestion) & C quantification
Tinker & Knight, 2000	Temperate‡‡	Wildfire	*50 (6.4)	Coarse down wood	Visual & volumetric
Eckmeier et al., 2007	Temperate§§	Experimental (slash-and-burn)	8.4 (5.4)	All (>1 mm)	Visual, gravimetric & total C quantification
Aponte et al., 2014	Temperate***	Prescribed (several)	n.d. (0.22–0.33)	Coarse down wood**	Visual, volumetric & total C quantification
Santín et al., 2012	Temperate***	Wildfire	n.d. (3.6–9.7)	Ash	Organic C quantification
Clay & Worrall, 2011	Temperate†††	Wildfire	4.3 (0.12)	Aboveground excl. standing wood	Visual, gravimetric & total C quantification
Worrall et al., 2013	Temperate†††	Prescribed	2.6 (n.d.)	Aboveground excl. standing wood	Visual, gravimetric & total C quantification
Fearnside et al., 1993	Tropical‡‡‡	Prescribed (slash-and-burn)	2.7 (n.d.)†	All but fine residues	Visual & gravimetric
Fearnside et al., 1999	Tropical‡‡‡	Prescribed (slash-and-burn)	2.9 (2.2)	All but fine residues	Visual & gravimetric
Fearnside et al., 2001	Tropical‡‡‡	Prescribed (slash-and-burn)	5.8 (3.2)	All but fine residues	Visual & gravimetric
Fearnside et al., 2007	Tropical‡‡‡	Prescribed (slash-and-burn)	4.0 (1.2)	All	Visual & gravimetric
Gráça et al., 1999	Tropical‡‡‡	Experimental (slash-and-burn)	8.4 (4.5)	All	Visual, gravimetric & total C quantification
Kauffman et al., 1995	Tropical‡‡‡	Prescribed (slash-and-burn)	1.4–5.3 (1.4–3.1)	Ash	Total C quantification
Carvalho et al., 2011	Tropical§§§	Prescribed	0–6.2 (n.d.)*	Experimental (wood blocks)	Visual & gravimetric
Kauffman et al., 1998	Tropical****	Prescribed (slash-and-burn)	3.7–4.8 (0.5–0.8)	Ash	Total C quantification
Righi et al., 2009	Tropical††††	Prescribed (slash-and-burn)	16.2 (6.0)	All	Visual, gravimetric & total C quantification
Rumpel et al., 2009;	Savanna‡‡‡‡	Prescribed (slash-and-burn)	0.4 (0.48)†	All	Visual, gravimetric & total C quantification
Kuhlbusch et al., 1996;	Savanna§§§§	Experimental (slash-and-burn)	0.6–1.5 (<0.05)	All	Thermo-chemical & C quantification
Saiz et al., 2014;	Savanna‡‡‡‡,§§§§	Experimental (1 m2 plots)	11–23 (0.23–1.24)†	Surface fuels excl. coarse DW	Visual & gravimetric& total C quantification
Donato et al., 2009;	Mediterranean‡‡	Wildfire	n.d. (0.3–0.6)	Down wood	Visual & volumetric
Goforth et al., 2005;	Mediterranean‡‡, §§	Wildfire	n.d. (1.15–1.45)	Ash	Organic C quantification
Alexis et al., 2007;	Mediterranean*****	Prescribed	5.4 (1.4)	All	Visual, gravimetric & total C quantification
Kuhlbusch & Crutzen, 1995	Various	Laboratory (burning apparatus)	0.1–3.4 (n.d.)	All fuels selected	Thermo-chemical & C quantification

Values given in relation to total mass consumed.

Values given in relation to prefire total C.

Values given in relation to woody fuel mass consumed.

Postfire sampling 1–2 year after fire without accounting for losses due to postfire erosion.

Postfire sampling 4–7 year after last fire.

Unless stated otherwise C quantification was not performed and C values (where reported) were obtained from previous studies.

Conifer fores.

Deciduous forest.

Eucalyptus forest.

Moorland.

Humid rainforest.

Seasonal conifer.

Seasonal grassland.

Seasonal semideciduous.

Grassland.

Open-tree grassland.

Scrub oak.

n.d.: not determined.

Firstly, most previous approaches have accounted for only some of the PyOM components formed in situ during fire. For instance, in the same forest complex investigated here, a 2% fuel mass conversion to PyOM was reported for a fire from the International Crown Fire Modelling Experiment (Lynch et al., 2004). However, that investigation only accounted for airborne PyOM collected in particle traps. It, therefore, excluded PyOM formed from, and remaining within, the forest floor, down wood and bark on standing trees. As another example, Fearnside et al. (2001) estimated a 6% PyC/CA conversion for a prescribed fire in the Amazonian rainforest, but they only accounted for woody charcoal pieces collected manually from the ground. They thus excluded, amongst others, the PyOM contained in fine residues (e.g., ash, charred forest floor or litter), which can be a substantial pool of PyC (Santín et al., 2012).

Secondly, the use of analytical approaches that quantify only a part of the PyOM continuum can lead to underestimation if the values obtained are assumed to represent the total PyOM produced. For example, Kuhlbusch et al. (1996) estimated a PyC/CA conversion rate of only 0.6–1.5% during an open-tree Savanna experimental fire, but this was done using a chemical/thermal oxidation method that quantifies only the most condensed forms of PyOM (Schmidt et al., 2001).

Thirdly, some studies used prescribed burns, which are usually not representative of wildfires, and sometimes carried out in human-manipulated fuels (Urbanski, 2014). For example, Fearnside, Gráça and collaborators have carried out several PyOM production inventories for tropical slash-and-burn fires (see details in Table3). Slash-and-burn fires are aimed at maximizing fuel consumption, resulting in high combustion completeness and, therefore, lower rates of PyOM production compared to wildfires (Kuhlbusch & Crutzen, 1995). It is worth highlighting here that, in their latest and most comprehensive study, they estimated a PyC/CA conversion rate of 16% (Righi et al., 2009), which is substantially higher than their earlier, and less complete, estimations (Table3: Fearnside et al., 1993, 1999, 2001, 2007; Gráça et al., 1999).

Our study overcomes the limitations outlined above as we (i) quantified PyOM produced in all fuel components; (ii) included the entire range of PyOM materials and (iii) examined a forest fire representative of typical wildfire conditions. To the authors’ knowledge, it therefore represents the most comprehensive quantification to date of PyOM produced in situ during a wildfire. Whilst drawing general conclusions from a single fire event, as that investigated here, must be done with caution, our results highlight a likely underestimation of PyOM production in previous studies of boreal forests and also beyond.

The boreal forest represents the world*s largest terrestrial biome and contains >30% of terrestrial C stock (Kelly et al., 2013), with wildfire being a dominant driver of the C balance here (Bond-Lamberty et al., 2007). Currently, 12.4 Mha of boreal regions burn on average each year and climate change is expected to lead to a substantial increase in wildfire season severity (Flannigan et al., 2013). There is already evidence that recent changes in climate have already lengthened the fire season in the North American boreal forest (Kelly et al., 2013). As our fire was typical for wildfires in the surrounding boreal region, it may be informative to scale up by combining the overall PyC/CA conversion rate from this study with global estimates for boreal regions of average fuel consumption (g C per m2 of area burnt; Van der Werf et al., 2010) and area burnt (Mha yr−1; Randerson et al., 2012). This gives a PyC production estimate within boreal regions of ∽100 Tg yr−1, which is more than five times higher than the previous estimates of 7–17 Tg yr−1 (Preston & Schmidt, 2006). Although this scaling up is rather speculative and the representativeness of the conversion rate found here for boreal forest fires needs to be validated more widely, this outcome suggests that boreal PyC production could represent a substantial C sink at the global scale.

The ability of PyOM to act as a C sink in the long-term is conditioned by its longevity in the environment. It is therefore important to recognize that, even if the overall resistance to degradation of PyOM is higher than its unburnt precursors (Schmidt et al., 2011) and some PyOM forms can persist in the environment for millennia, others can be mineralized relatively fast (within days or months) (Singh et al., 2012; Zimmerman et al., 2012). The longevity of PyOM depends on both its intrinsic resistance to degradation (mainly driven by fuel properties and burning conditions, e.g., Soucémarianadin et al., 2013) and on the characteristics of the environment itself (e.g., oxygen availability, physical protection; Marschner et al., 2008; Singh et al., 2014). Therefore, for a complete quantitative assessment of the C sequestration potential of wildfire PyOM, an evaluation of the resistance to degradation of the different forms of PyOM is also required. In boreal regions, a particularly long half-life in the range of hundreds or thousands of years can be expected for the most recalcitrant PyOM fractions, because of specific formation and environmental conditions (e.g. relatively high production temperatures and cool climate) (Preston & Schmidt, 2006).

Given the overall importance of the boreal biome in the global C balance, the finding that nearly a third of the C in boreal biomass affected by wildfire could be transformed into PyOM rather than emitted to the atmosphere is significant. The inclusion of wildfire PyOM production, and its C sequestration potential, in C budgets would, therefore, be an important step in reducing the uncertainty in C accounting and, thus, future climate projections (Lehmann et al., 2008). At present, C cycle uncertainties are among the major unknowns affecting scenario development (Moss et al., 2010) and wildfires are one of the environmental perturbations least understood in terms of their impact on the global C cycle (Reichstein et al., 2013). Although our study is focused on the boreal region, we have identified a previous general underestimation of PyOM production from wildfire studies that applies, to some extent, also to other major fire-prone ecosystems (Table3). Further research, including the comprehensive quantification of PyOM production for a range of ecosystems and fire behaviours, as well as the longevity in the environment of the different forms of PyOM, is required to fully address the role of PyOM in the global C budget.