Fire-derived organic carbon in soil turns over on a centennial scale

: Pyrogenic carbon (PyC), the residue of an incomplete combustion of biomass, is considered as a carbon (C) sink due to its assumed stability in soil. PyC turnover time estimated using two modelling approaches, based on data from 16 published studies (n = 54) on PyC degradation, ranged from a decadal to centennial time scale, varying with initial biomass type, pyrolysis temperature, and incubation or field study. The average turnover time using a one-pool approach was 88 y, and the best estimate using a two-pool approach was 3 y for a fast-cycling pool and 870 y for a slow-cycling pool. Based on this meta-analysis, PyC cannot be assumed to persist in soils for thousands of years, and its use as a strategy for offsetting carbon emissions requires prudence and further research. Abstract Pyrogenic carbon (PyC), the residue of an incomplete combustion of biomass, is considered as a carbon (C) sink due to its assumed stability in soil. PyC turnover time estimated using two modelling approaches, on data from 16 published studies (n = 54) on PyC degradation, ranged from decadal to centennial time scale, varying with initial biomass type, pyrolysis temperature, and incubation or field study. The average turnover time using a one-pool approach was 88 y, and best estimate using a two-pool approach was 3 y for a fast-cycling pool and 870 y for a slow-cycling pool. Based on this meta-analysis, PyC cannot be assumed to persist in soils for thousands of years, and its use as a strategy for offsetting carbon emissions requires prudence and further research.


Introduction
Wildfires transfer approximately 0.05 to 0.2 Pg C yr -1 to soil (Seiler and Crutzen, 1980;Kuhlbusch, 1998) as incomplete combustion residue of biomass, known as pyrogenic carbon (PyC) (Goldberg, 1985). Climate change is projected to increase wildfire frequency in many parts of the world (Flannigan et al., 2006), which could modify the input of PyC and consequently the terrestrial carbon cycle (Westerling et al., 2006). PyC is ubiquitous in the environment and ranges from 2% to 45% of the total soil organic carbon (SOC) in terrestrial systems (Bird et al., 1999;Schmidt et al., 1999;Lehmann et al., 2008). Some researchers suggest that PyC forms a slow-cycling C pool in the soil (Preston and Schmidt, 2006;Marschner, 2008). If so, conversion of plant biomass to PyC would represent a transfer of faster-cycling biomass-C to slower-cycling C in soils (Ohlson et al., 2009) and is therefore expected to act as a C sink (Seifritz, 1993;Marris, 2006). In the last decade, PyC has gained interest as a strategy for sequestering atmospheric CO 2 to partly offset carbon emissions (Lehmann et al., 2006).
The ability of PyC to act as a carbon sink depends on its persistence in the soil. PyC is widely considered to be relatively "inert" (Forbes et al., 2006) because PyC has been preserved in geological samples or strata (Forbes et al., 2006), archaeological sites (Schmid et al., 2002;Glaser, 2007), and old anthropogenic soils (Glaser et al., 2000; Knicker, 2011; Glaser and Birk, 2012). Moreover, in some experiments, PyC was resistant to chemical oxidants (Skjemstad et al., 1996) and contributed to the oldest soil organic carbon (SOC) pool in some Australian soils (Krull et al., 2006). Based on the 14 C age of PyC macro-pieces/charcoal (Pessenda et al., 2001;Schmidt et al., 2002) and budget calculations (Forbes et al., 2006), PyC age in soil has been estimated to be on the scale of hundreds to ten thousand years (Liang et al., 2008). The limitation of using radiocarbon age to estimate turnover time is that we rarely have knowledge of the input rate (or, for isolated systems, initial stock when the radiocarbon "clock" started), which would be needed to estimate turnover times.  Schmidt, 2009) and chemically by abiotic (Lehmann et al., 2005;Cheng et al., 2006;Hockaday et al., 2006) and/or microbial agents (Potter, 1908;Shneour, 1966;Goldberg, 1985). Incubations have identified abiotic (Cheng et al., 2006) and biotic oxidation processes (Potter, 1908;Hamer et al., 2004;Kuzyakov et al., 2009;Zimmerman, 2010) as important mechanisms of PyC degradation. Turnover times of PyC reported in most of these experimental studies ranged between a hundred and a thousand years.
These recent observations contradict the perception that PyC persists in soil for millenia.
The uncertainty in PyC persistence is accompanied by a basic lack of understanding about PyC dynamics in soil. Spokas (Spokas, 2010) observed an increase in stability of PyC with decrease in O:C molar ratio of PyC. However, the correlation between half life of PyC with O:C molar ratio was based on different methodological approaches to estimate mean residence time of PyC. Therefore, to reconcile the apparent discrepancies between assumed persistence of PyC based on radiocarbon age and fairly rapid degradation of PyC as observed in experiments, we assembled data from published studies on PyC losses from soil and, for the first time, calculated turnover times within and across all studies with one consistent approach.

Data set collection from the literature
We compiled data from published studies (n = 54 data sets from 16 studies, supplement Table 2) on PyC degradation. We investigated turnover times of PyC using two previously published models to describe PyC decomposition and/or soil organic matter dynamics (Supplement Table 1). These models should be seen as a way of approximating characteristic time constants, rather than quantifying the exact dynamics (Burnham and Anderson, 2002).

One-pool approach
In the first approach, we used a one-pool exponential decay model in which PyC is modelled as a single homogeneous C pool and assumed to follow first order kinetics (Brodowski, 2005;Cheng et al., 2008b;Hammes et al., 2008b;Nguyen et al., 2008). We assumed that there were no new PyC inputs between time = 0 and time = t (in years). We calculated the decay rate from the total loss of PyC (sum of all loss processes including leaching, erosion, mineralization, and/or decomposition) relative to the initial stock, to estimate the turnover time of PyC in the soil with respect to all loss pathways of PyC from the soil to other terrestrial pools or from the terrestrial ecosystem.
Based on these assumptions, the decay rate k is calculated from the loss of PyC over time as follows: where C t is the remaining stock after time t, C 0 is the initial stock of PyC (at t = 0), k is the decay rate (y -1 ). The turnover time τ is calculated as τ =1/k.
For the one-pool approach, we calculated turnover times based on two data points for each study, the initial stock of PyC and final PyC remaining at the end of the experiment,

Two-pool approach
In the second approach, PyC decomposition dynamics were calculated using a two-pool was characterized by slow decay rate, k slow . We assumed that the pools decayed in parallel-in other words that there was no exchange of PyC between pools. Thus, where, C t is the remaining stock after time t; x is the proportion of initial stock in the fast cycling PyC pool (at t = 0), C fast ; (1-x) is the proportion of slow cycling pool (at t = 0), C slow ; k fast and k slow are decay rate constants (year -1 ).
Accordingly, the turnover time for the fast cycling pool τ fast (y) is 1/k fast and for the slow cycling pool τ slow (y) is 1/k slow .
The two-pool model was fitted to the compiled data set (n = 54) (with the initial stock at time t = 0 and stock at the last point for each study corresponding to the time series decrease in initial stock with time) using the constrained non-linear parameter estimation procedures in the IBM SPSS statistics software package for the Mac. The curve-fitting values were iterative and required initial starting values. To avoid errors due to convergence to local minima of residual sum of squares (RSS), we adopted convergence criteria as used by Updegraff (Updegraff et al., 1995), where final parameter estimates were accepted only if equations converged to the same values given starting values up to 50% above and below them. The explained variance for the two-pool model is given in Supplement Table 3.

Assumptions
First, for the one-pool decay model, we accepted the simplification to one homogeneous Therefore, the data set is restricted to those studies where (1) the initial inputs and stock were known or could be estimated; (2) the initial stock decreased or remained constant with time; and (3) the experimental setup included terrestrial systems.

Turnover time of PyC for combined dataset
The turnover time computed for each study using Eq. (1) range from <1 to 750 y and yielded an average value of 88 y (with standard deviation as 131 y and standard error of mean as 18). The large standard error represents a large variation in the experimental studies. The overall turnover time of PyC computed with the one-pool decay model using model fit by non-linear regression and chi square minimization was 291 y (r 2 = 0.32, n=54, root mean square error = 10.13). The turnover time computed with the two-pool model was 3 y for the fast cycling pool (C fast =17%) and 870 y for the slow cycling pool (C slow = 83%) of PyC (r 2 = 0.44, root mean square error = 8.35). The two-pool model gave a slightly better fit to the data than did the one-pool model (Fig. 1).
The calculated turnover times are much shorter than previously assumed or estimated to date. The higher number of short-term studies in the compiled data set, which mainly capture the fast cycling dynamics, could influence the overall calculated turnover time to a faster value. Although having faster decay than previously thought, the overall turnover times suggest that PyC is more stable than all known plant-derived organic compound classes in soil, based on low level 13 C labelling experiments (Amelung et al., 2008; Glaser, 2005).

Turnover time of PyC as a function of different factors
We observed a high scatter in the turnover times between different studies (ranged from <1 to 750 years) (Fig. 2) We grouped the data to see whether these factors influenced turnover times when all other factors were allowed to vary, namely (1) incubation vs. field studies; (2) type of biomass (grass vs. wood); (3) pyrolysis temperature (<400ºC and ≥ 400ºC); and (4) quartz sand vs. soil medium (Fig. 3). Data were not grouped by other factors that control SOM decomposition-like climate, degree of soil development, soil types, topography, and biota -because they were either not reported or were kept constant in most studies.
We computed individual turnover times for each data set (n=54) using Eq. (1) and computed the range, average, and variation of turnover times associated with each of the above-mentioned factors. To avoid the effect of the differences due to time scale, we only considered incubation studies and therefore used Eq. (1) for the grouped data. For comparison between incubation studies and long-term field studies, we also computed turnover time using Eq. (2) for the long-term field study to take into account the slowing down of mineralization rate with time. The turnover times of grouped factors were compared using a non-parametric Wilcoxon rank sum test. Interactions between factors on the compiled data were evaluated by multi-way ANOVA using R software (Supplementary Table 3). Our analysis shows that we do not have any significant interactions between the factors. Therefore, the unbalanced design of the grouped data does not introduce any significant error in the interpretation, and we can evaluate the differences in turnover times associated with these different factors.

Incubation vs. Field study
Incubation studies have significantly (p<0.001) shorter turnover times (average 55 y; range 1-180 y) than field-based studies (average 353 years; ranges from 90 to 750 years) computed using the one-pool decay model, Eq. (1) (Fig 3). Short-term decay studies may primarily capture the rapid initial loss of more labile or/and non-charred components, and therefore may not be a good indication of the long-term degradation rates. The two-pool model partly solves this problem, but we had insufficient information to parameterize this Long-term field studies provide more realistic estimation of in situ turnover time of PyC, which not only includes the rapid initial phase but also the phase when mineralization rate decreases with time. We took advantage of published data from long-term field studies to estimate the turnover time for PyC in situ. It was, however, not possible to conduct a straightforward mass balance for the multi-year field studies because few of these had data on initial PyC stock and the rate of atmospheric deposition of PyC throughout each study was unknown. The turnover time derived by analysing all the data from long-term field studies (n=6) using the two-pool model were 91 y for the fast cycling pool and 1034 y for the slow cycling pool, with 49% in the fast pool (r 2 = 0.51) (Supplement Fig. 1). Thus, long-term field data indicate that a significant fraction of PyC turn over on roughly centennial scale, which is shorter than previously assumed or

Initial biomass type
There were two types of initial biomass in the studies we used that were representative for grassland and forest ecosystems, namely grass and wood (Fig. 3) Therefore, the chemical and physical structure of PyC is not directly correlated with or predicable by the structure of the plant substrate for charring.

Pyrolysis temperature
The pyrolysis condition under which PyC is formed also determines its chemical and physical properties and possibly its turnover times. In natural environments, it is unlikely that any one set of formation conditions can be viewed as typical (

Quartz sand and soil medium
Most incubation studies used either quartz sand with microbial inoculum or fresh soil, and all field studies took place in soil. As a consequence, we only used incubation studies However, short-term incubation studies might not capture the stabilizing effects of organo-mineral interactions.

Climate
Climate, including temperature and moisture, influences SOM and PyC decomposition, but there were insufficient data with which we could analyse its effect on turnover time. First, the apparent inconsistency of PyC radiocarbon ages with the estimated turnover time of PyC could be explained by the ''inbuilt age'' of a piece of charcoal produced during fire because the wood may have been old at the time of the fire (Gavin et al., 2003). Few trees live to be a thousand years old, so it is likely that a piece of charcoal that is thousands of years old has been in the soil for >1000 years. Moreover, it is difficult to translate the radiocarbon age of an isolated piece of charcoal to a turnover time without knowledge of the initial stock or input of PyC, leading to uncertainty in the estimate.
Second, it is likely that some PyC stays in some soils for many thousands of years.
The two-pool model shows that the PyC experiments analysed contained fairly slow cycling material, even if the bulk behaviour was well described with a shorter turnover time.
Studies of bulk soil organic matter find a spectrum of turnover times, with persistence depending on the compound chemistry and its physico-chemical state in soil, such as interaction with minerals or protection inside aggregate structures (Schmidt et al., 2011) that result in turnover times up to thousands of years. We would expect PyC to have similar behaviour in soil (Torn et al., 2002).
A caveat on our turnover time results is that at present, most controlled studies of PyC are relatively short term and may be biased towards rapid turnover times, given the initial decomposition dynamics that diminish over time (Kuzyakov et al., 2009) as the labile component is metabolized (Smith et al., 2010). Our knowledge of the later stages of degradation is much poorer. Therefore, our estimates using a one-pool model could overestimate the rate of PyC degradation (and underestimate turnover time). The twopool decay model provides a better fit to the data than the one-pool model, and does show a more persistent fraction of PyC. It has been suggested that physical protection and interactions with soil minerals play a significant part in long-term PyC stability (Brodowski et al., 2006;Glaser et al., 2000) and forming what we called the slow PyC pool. However, the duration of available data is too short to quantify longer decay time scales.
The differences in turnover times among studies could also be due to the interplay of different decomposition or stabilization mechanisms at a local scale, resulting in differences in the rate of degradation. A combination of physical, chemical, and microbial processes can play a role in PyC degradation. However, these degradation processes have not yet been studied in combination, and questions remain as to the interaction of these processes and the importance of PyC chemistry, soil conditions, and microbial activity in controlling the likelihood of PyC degradation or persistence.

Conclusion and future research
PyC comprises an array of compounds and is present in different environmental matrices;   327  328  329  330  331  332  333  334  335  336  337  338  339  340  341  342  343  344  345  346  347  348  349  350  351  352  353  354 355 Fig. 1: Parameter estimates for the one-pool exponential and the two-pool exponential models. Turnover time calculated using first order decay vs. duration of experiment (left). The empty symbol represents incubation studies, 362 filled symbol represents field based studies, circles correspond to grass PyC, squares corresponds to wood PyC, colour represents 363 sand (blue) and soil (red) medium, the numbers represent temperature of pyrolysis: (1) for <400oC and (2) for ≥ 400oC. There is a 364 weak relation between experiment duration and individual turnover time (r 2 = 0.49), showing experiment duration is not the only 365 factor influencing turnover time. Box plot of individual turnover time for each study (right), where filled black circles are outliers 366 beyond 5th or 95th percentiles. 367 368 369