Interactive comment on “ How past fire disturbances have contributed to the current carbon balance of boreal ecosystems ? ”

Introduction Conclusions References


Introduction
Boreal vegetation covers about 17 % of the Earth's land surface but contains more than 30 % of all terrestrial carbon stocks (Kasischke, 2000).This above average carbon density reflects the large amount of soil organic carbon being conserved thanks to the Figures

Back Close
Full general cold and wet soil conditions, especially in peat and carbon-rich frozen soils (Harden et al., 1992;Jones and Yu, 2010;Tarnocai et al., 2009).Mainly in response to increasing atmospheric CO 2 and climate change, boreal ecosystems are estimated to be a net carbon sink for the past two decades (Kurz and Apps, 1999;McGuire et al., 2009;Pan et al., 2011b).Yet, as climate change continues, boreal forest may become more vulnerable, as indicated by (1) deceleration of "greening" over this biome as seen by satellites (Xu et al., 2013), (2) locally observed decreased vegetation activity (Beck and Goetz, 2011), and (3) evidence for large climate-related disturbances such as insect outbreaks (Kurz et al., 2008) and catastrophic fires (Kasischke and Hoy, 2012) that cause CO 2 losses to the atmosphere.
Fire has always been a natural disturbance in boreal ecosystems (Anderson et al., 2006), and it has multiple impacts on vegetation dynamics, carbon cycling, soil processes, atmospheric chemistry and permafrost dynamics.Fire plays an important role in the evolution of ecosystem species composition in this region through complex fireclimate-vegetation feedbacks at different time scales (Kelly et al., 2013;Schulze et al., 2012).The carbon balance of boreal forest is modified immediately by fire through firecarbon emissions, but fires also lead to successional post-fire carbon accumulation as the ecosystem recovers -a long-term process of CO 2 removal from the atmosphere (Amiro et al., 2010;Goulden et al., 2011).Besides, fires impact soil carbon dynamics, primarily by direct combustion of the organic layer at the soil surface, but also through the creation and deposition of recalcitrant charcoal and restoring soil carbon to equilibrium in parallel with the post-fire ecosystem recovery (Santín et al., 2015).Further, soil carbon dynamics are also changed by altered soil temperature and moisture conditions after fire (Harden et al., 2006).
Many factors contribute to the currently observed boreal carbon sink, including: the fertilization effect of increasing CO 2 concentration (Balshi et al., 2007), nitrogen deposition (DeLuca et al., 2008), forest management (Kauppi et al., 2010), climate change (Wang et al., 2011), and the balance between ecosystem (mainly forest) recovery from past disturbances (Pan et al., 2011b) and emissions from current fires.However, the Introduction

Conclusions References
Tables Figures

Back Close
Full relative contributions of these factors and their interactions are still poorly known, although a large part of the carbon sink in boreal forests has been attributed to forest recovering from past disturbance or degradation (Kauppi et al., 2010;Pan et al., 2011a).
Given the role of fire in driving the demography and carbon balance of boreal forests, several studies used biogeochemical models to examine the carbon balance of boreal ecosystems and the related impacts by fires (Balshi et al., 2007;Hayes et al., 2011;Yuan et al., 2012).These studies conducted simulations with fire and without fire (or with stationary fire regime) and examined the total sum impacts of all preceding fires on the boreal carbon balance for a particular "target" time period.However, the immediate source impacts of current fires through emissions and the sink legacies by previous fires were not formally separated.Consequently, the contributions of fires that occurred before the contemporary period (and associated post-fire vegetation recovery) to the current carbon balance, i.e., the legacy sink effects of past fire, remained largely unknown.
Here, we develop a conceptual framework to quantify the decadal contributions of 2 Materials and methods

Model introduction
This study uses the process-based dynamic global vegetation model (DGVM) OR-CHIDEE (Krinner et al., 2005).The ORCHIDEE model has three sub-modules.The SECHIBA sub-module simulates the fast exchange of water and energy between the land and the atmosphere.The STOMATE sub-module simulates the vegetation carbon cycle processes including: photosynthesis, photosynthate allocation, litter fall, litter and soil organic matter decomposition.The third sub-module simulates vegetation dynamics.The equations of vegetation dynamics are mainly taken from the LPJ model (Sitch et al., 2003), with modifications being described by Krinner et al. (2005).
For this study, the prognostic fire module SPITFIRE as originally developed by Thonicke et al. ( 2010) was incorporated into ORCHIDEE, from here on referred to as ORCHIDEE-SPITFIRE.Global validation of simulated burned area and firecarbon emissions were described by Yue et al. (2014) and Yue et al. (2015).Notably, ORCHIDEE-SPITFIRE is able to capture the decadal variations of burned area in boreal Russia when compared with the historical reconstruction data by Mouillot and Field (2005), and the interannual variations of burned area in boreal North America when compared with the fire agency data.All fire processes are the same as described in Yue et al. (2014), except that the suppression of lightning-ignited fires by human is introduced, as a function of human population density, following Li et al. (2012): where, D p is the population density (individuals per km 2 ), and F s a multiplicative coefficient applied to lightning ignitions to account for human suppression at a given D p .This corresponds to a suppression fraction of 0.01 in sparsely inhabited regions and of 0.99 in highly populated regions (i.e., D p → +∞).
Within SPITFIRE, fire occurrence depends on vegetation and climate conditions, and has feedbacks on forest mortality through crown scorching and cambial damage, which Figures

Back Close
Full reduces forest stem density (Thonicke et al., 2010).Thus in ORCHIDEE-SPITFIRE, vegetation dynamics are affected by both climatic factors, as simulated by the dynamic vegetation module, and fire disturbances as simulated by SPITFIRE.On top of the climatic limits that give the adaptation or extinction for different tree vegetation types under specific climate and climate variability conditions (Krinner et al., 2005;Sitch et al., 2003), fires further impact the tree-grassland competition and the competition within woody vegetation types.
The ORCHIDEE-SPITFIRE used here includes the DGVM improvements made by Zhu et al. (2015), which improved the simulation of northern vegetation distribution.The improved DGVM processes include: (1) tree mortality dependence on growth efficiency, defined as the ratio of net annual biomass increment to the preceding-year maximum leaf area index (LAI); (2) tree mortality induced by winter extreme coldness for all tree plant functional types (PFTs) except boreal deciduous needleleaf, and by spring frost in broadleaf forests only; (3) definition of the treeline limit to be an isotherm of growing-season mean soil temperature of 6.7 • C. A threshold of mean monthly temperature of 22 • C is used to limit the distribution of C4 grass, following Still et al. (2003).Maximum carboxylation rates (V cmax , µmol m −2 s −1 ) were adjusted based on the results of parameter optimization for ORCHIDEE against flux tower measurements (Kuppel, 2012).

The conceptual framework
In this section we develop a conceptual framework which forms the basis of our sim- nitrogen deposition, hereafter referred to as the CCN perturbation (note this term is intended to include other environmental changes that are not explicitly included in our discussion here, for example, radiation change and aerosol effect).
Typically, natural land ecosystems in equilibrium with fire under pre-industrial conditions (no CCN) are expected to be carbon neutral at a large scale, i.e., having a carbon balance close to zero.Stand-replacing fire disturbances break this steady state locally, releasing CO 2 and initiating forest regrwth (Amiro et al., 2010;Goulden et al., 2011;Odum, 1969).At steady state over a large region, the flux of CO 2 lost to the atmosphere by fires affecting some forests, is offset by the space-time integral of legacy sinks in other forests regrowing from previous fires (Fig. 1b black curve).However, the CCN perturbation affects both fire-carbon emissions (e.g.drier summers may increase fires) and legacy sinks (e.g. higher CO 2 accelerating regrowth).For instance, considering the decade of 2000-2009, the carbon balance of a grid-cell is the sum of (1) fire emissions during 2000-2009, (2) legacy sink caused by fires that occured since 1850 and impacted by CCN to various degrees (shown as the blue curve in Fig. 1b), and (3) source or sink of the tracts of forests that have not burned since 1850 but are influenced by CCN.The compositon of the 2000s-decade carbon balance is illustrated in Fig. 1a.
Similar to the attribution framework for land use change carbon fluxes established by Gasser and Ciais (2013), the carbon balance of a geographical area covered by a given biome (g, b) for the 2000s decade, under the CCN perturbation and taking into account decadal fire disturbances since 1850, can be expressed as: where (2), we separated the total carbon flux into lands undisturbed and those disturbed by fire.Further, we assume that fires also occurred before 1850 but their influence on the 2000s-decade carbon flux are included in the undisturbed land flux, given the observed very small net ecosystem productivity in boreal forests older than 150 years old (Goulden et al., 2011).
In studies using numerical biogeochemical models, Eq. ( 2) represents a case in which fire-generated forest cohorts are explicitly simulated -the 2nd part on the right hand of the equation gives the contributions of different decadal fires to the carbon balance for 2000s decade.However, for models that do not explicitly simulate forest cohorts, a workaround is possible by manually suppressing fires in the model within some particular decade, to allow quantifying the contribution of fires from this decade by the difference between the two simulations.Similar as Eq. ( 2), the carbon flux for the 2000s decade in case fires are suppressed in some particular decade D could be written as: where where Cont D is the contribution of fires within the D decade to the carbon balance of the 2000s decade.Different with explicit cohort simulation, this factorial approach quantifies the past-fire-generated "cohort" contribution taking as a baseline the carbon flux of otherwise undisturbed land but as influenced by the CCN perturbation.Finally, one could vary D from 1850s to 2000s to derive the contribution by fires within each decade between 1850-2009.This conceptual framework remains valid when integrating all the variables in Eqs. ( 2)-( 4) over the geographical extent and different vegetation types to attribute carbon fluxes at regional scale.Note in this framework, different factors of the CCN perturbation are not seperated and the CCN perturbation impact is embedded in the fire contribution.

Simulation protocol and input data sets
Following the conceptual framework, we conducted factorial simulations to quantify the decadal contributions of past "fire cohorts" to the simulated carbon balance of 2000-2009.The carbon balance is defined as the Net Biome Production (NBP): where NPP is net primary production (i.e., the net biomass accumulation by plants after accounting for their own use), RH is the ecosystem heterotrophic respiration, EMI is carbon released by fire.A positive NBP indicates a net carbon flux from the atmosphere to land, i.e., a land carbon sink.In the following, we use the terms "carbon sink" and "NBP" interchangeably, unless otherwise specified, i.e., that a negative NBP is a carbon source releasing carbon to the atmosphere.We fire disturbance.We then conducted a series of other simulations (named SIM OFF ) which branch off from the SIM fireON simulation from the beginning year of each decade between 1850 and 2009.In the SIM OFF simulations, the fire module was switched off sequentially from the decade of 1850s (1850-1859) to 2000s (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009) and switched on afterwards, with all remaining parameter settings and input data sets the same as in the reference simulation.Following the Eq. ( 4), the contribution by fires within some specific decade to the carbon balance of each year for the time after this decade would be quantified as the difference between the reference simulation and the decadal SIM OFF simulation.In all simulations, the vegetation dynamics module of ORCHIDEE was switched on to allow the vegetation distribution to respond to climate variations and fire disturbances.
The spatial domain of our simulation covers the land pixels of 44-84 • N at 2 • resolution.The land north of 84 • was excluded as it is covered mainly by ice and snow.As our focus is carbon dynamics of natural vegetation in response to fires within the boreal region, croplands were not simulated in the model.This is acceptable given that land-use change during the 20th century in this region was small (Hurtt et al., 2006).Cropland fractions within grid cells were prescribed according to a current-day vegetation map (the IGBP-DIS 1 km global land-cover map, Loveland et al., 2000), and fractions of natural vegetation (i.e., trees and grasses) were simulated.Tundra in the high-arctic regions is simulated as C3 grassland.

Comparison of simulated forest distribution and fires to observations
We compared the spatial distribution of three morphological and phenological tree groups between the model simulation and MODIS land-cover data for the year 2010: broadleaf (including evergreen and deciduous), evergreen needleleaf and deciduous needleleaf trees, corresponding to the three boreal tree PFTs in ORCHIDEE.The MCD12Q1 version 5 land-cover data (Friedl et al., 2010) were used (http:glcf.umd.edu/data/lc, with a northern limit of 84 • N).Fractions of the 17 different land-cover types in the IGBP land classification scheme were calculated at a 2-degree resolution based on the 500 m original resolution data.Further, the 2-degree land-cover fractions were cross-walked to PFT fractions using the approach developed by Poulter et al. (2011), in which the mixed tree-grass land-cover types such as shrublands are assumed to be Introduction

Conclusions References
Tables Figures

Back Close
Full composed of different fractions of trees and grasses (see Table 6 in Poulter et al., 2011 for more details).The simulated maximum foliage projective cover for each of the three tree groups was compared with the corresponding MODIS observation, with the sum of the three groups being compared as tree cover.Simulated burned area and fire-carbon emissions were compared with GFED3.1 burned area data (Giglio et al., 2010) and carbon emission estimates simulated by the CASA biosphere model (van der Werf et al., 2010).Burned areas and fire-carbon emissions from agricultural fires were excluded from GFED3.1 data before comparison, because these fires are not included in the model.Northern peatland fires were not simulated due to a lack of peatland PFT in the model, nor are they included in the GFED3.1 emission data.

Simulated forest distribution
The simulated spatial extent of forest distribution is broadly similar to that of MODIS land cover data over the region north of 44 • N for year 2010, with the forest biome extending from eastern Canada northwestward to Alaska in boreal North America, and that in northern and northeastern Europe, as well as most of Siberia (Fig. 2).The magnitude of foliage projective tree cover between ORCHIDEE and MODIS land-cover data is generally comparable, except in the southern and northern fringes of the study region (mainly Asia and America), where tree cover is overestimated by approximately 30-50 % in ORCHIDEE (hatched areas in Fig. 2). Figure 3 presents simulated and observed spatial distribution of three tree groups: broadleaf (including evergreen and deciduous), evergreen needleleaf and deciduous needleleaf.There is a widespread presence of broadleaf forest but of general low fractional cover across the study region, which is fairly reproduced by ORCHIDEE (Fig. 3  the dominance of evergreen needleleaf forest in North America, and in western Siberia and northern and eastern Europe (Fig. 3 panels 2a and b).In contrast, MODIS data show that central and eastern Siberia is dominated by deciduous needleleaf forests (Fig. 3 panel b).ORCHIDEE successfully captures this, but the spatial extent and magnitude of tree cover are overestimated (Fig. 3 panel a).In addition, ORCHIDEE also erroneously allocates more deciduous needleleaf forests in Alaska and northwestern Canada than the MODIS data.

Simulated burned area and fire-carbon emissions
The spatial distribution of simulated mean annual burned fraction for 1997-2009 is compared with GFED3.1 data in Fig. 4, with non-modelled agricultural fires being excluded from GFED data.The comparisons of cumulative latitudinal distribution of burned area and fire-carbon emissions are shown in Fig. 5.Although spatial disagreements in burned area exist, ORCHIDEE-SPITFIRE simulates an annual total burned area of 11.9 Mha yr −1 and fire-carbon emissions of 0.20 Pg C yr −1 , which are close to GFED3.1 estimates giving an annual burned area of 16.9 Mh yr −1 and fire-carbon emissions of 0.20 Pg C yr −1 .Spatially, burned area is underestimated within the latitude band 44-54 • N in Eurasia, concurrent with an overestimation of tree cover in the same region (Figs. 2 and 3).On the other hand, there is an overestimation of burned area in the regions north of 54 • N covered by forest, shrubland and tundra according to the MCD12Q1 land-cover map.Over North America, the spatial distribution of simulated burned area is in fair agreement with the GFED3.1 data, with burned area being dominated by the northwest-to-southeast boreal forest fires.

Decadal contributions of fire to the simulated carbon sink
The ) as assessed by IPCC AR5 (Ciais et al., 2013).Figure 6b shows how each decadal fire cohort contributes to the NBP of the study domain.For example, the curve labelled "1910s" shows the annual contribution of the 1910s-decade cohort, which produced a net carbon source during 1910-1919, followed by a long-term carbon sink whose magnitude decreases with time.Note that for the decade of 2000s, all fires before this decade contribute as a carbon sink term with varying sink sizes, whereas fires within the 2000s decade contribute as a source term.The whole study region can be classified into six fire groups according to their different fire return intervals (FRIs, here quantified as the inverse of burned fraction) as simulated by the model, with the shortest FRI of 2-10 yr and the longest of more than 500 yr.This classification was done for each decade of 1850-1999 (i.e., decades hav-Introduction

Conclusions References
Tables Figures

Back Close
Full ing a carbon sink effect for 2000-2009) using simulated mean decadal burned fraction, followed by partitioning decadal sink contribution into these fire groups.Figure 8 shows relative contributions of each fire group by summing together the partitioning results of all the decades.The fire group with an FRI of 10-50 yr emerges as the biggest contributor, contributing a carbon sink of 0.1 Pg C yr −1 or 42.7 % of the total sink effect.are natural late-successional species, frequent surface fires are the major factor explaining the dominance of Larix over the evergreen climax tree species.Infrequent crown fires initiate new Larix cohorts while surface fires thin them and prevent evergreen needleleaf saplings from reaching the canopy.Even though our model does not account explicitly for these two different fire impacts, over a broad scale, the dominance of evergreen coniferous forests in northern Europe and western Siberia coincides with slightly lower fire frequencies (Figs. 3 and 4).This is consistent with the observed pattern that more frequent fires in eastern Siberia are associated with the dominance of Larix deciduous needleleaf trees.
For the majority of the pan-boreal region, ORCHIDEE-SPITFIRE simulates a fire return interval of 10-200 years (Fig. 4, corresponding to burned fraction of 0.5-10 %), which is consistent with the evidence from various observational data sets (Giglio et al., 2010;Stocks et al., 2003).The simulated fire frequency (0.2-2 % yr −1 ) in Canada agrees with that reported by Stocks et al. (2003) using the Canadian Large Fire Database.The general spatial extent and magnitude of fires in northern Eurasia (> 54 • N) roughly agrees with GFED3.1 data, although burned fractions in northern tundra and shrubland are overestimated.This might be because tundra is treated as generic C3 grass in the model and thus assigned a low fuel bulk density (Thonicke et al., 2010) that promotes fast fire propagation.In reality tundra has a more dense growth form than temperate grasslands and therefore has a much higher bulk density (Pfeiffer et al., 2013).Fires are greatly underestimated by the model at the southern edge of the study area in Eurasia, with a simulated burned fraction of 0.2-2 % compared to values of 1-30 % in GFED3.1 data.This underestimation, especially in central Asian grasslands over Kazakhstan and Mongolia, is accompanied by an overestimation of tree cover (Fig. 2).This indicates that the role of fires to promote grasslands against forests as shown by other modelling studies (e.g., Bond et al., 2005;Poulter et al., 2015) in these semi-arid regions is underestimated in ORCHIDEE-SPITFIRE, probably due to excessive tree sapling recruitment.Figures

Back Close
Full

Role of fires in regional carbon balance and comparison with other studies
Consistent with the fact that fires are a large source of CO 2 at the time of burning, we found that fires during 2000-2009 emitted 0.18 Pg C yr −1 , close to its carbon source contribution (−0.17 Pg C yr −1 ) to the 2000s-decadal carbon balance.However, this source effect is compensated by legacy sinks in lands recovering from fires prior to 2000s, which are ameliorated by climate warming and CO 2 fertilization.Using factorial simulations, we quantified the relative sink contributions of fires in different decades of the past and further found that more than 60 % of the sink effects are contributed by fires during 1960-1999.This is a feature that differs our study from a few previous modelling studies in boreal ecosystems that also examined the role of fires in regional carbon balance (Balshi et al., 2007;Hayes et al., 2011;Yuan et al., 2012).Balshi et al. (2007) and Hayes et al. (2011) used additive simulation protocol to examine fire impact on the carbon balance, i.e., the contribution of fire to the carbon balance of some "target" decade (e.g., 2000s) is given by the difference between two simulations, with and without fires, respectively.Note that this approach examines the total sum effect of all fires occurring before but also within the target decade, i.e., equivalent to the effect of all fires of 1850-2009 termed as "net fire effect" in our analysis.Balshi et al. (2007) further conducted parallel simulations with and without CO 2 fertilization for all additive runs.They found that during 1996-2002, the sum effect of fires in the pan-boreal region (north of 45 • N) increased the ecosystem carbon storage (ranging 0.08 to 0.5 Pg C yr −1 ) for all years except 2002, according to a simulation that includes the CO 2 fertilization effect.When CO 2 fertilization effect is excluded, the role of fires is more varied, leading to an almost close to zero sum fire effect for the same period.We also found the "net fire effect" during the 2000s decade to be a carbon sink of 0.06 Pg C yr −1 (i.e., equivalent to the sum fire effect in Balshi et al., 2007), being smaller than that reported in their study.However, we noticed that in their study the contribution of fires varied greatly in magnitude from year to year, and sometimes even three times Figures

Back Close
Full higher than the sink term by the CO 2 fertilization effect, which may indicate the great uncertainty in their results (Fig. 6 in Balshi et al., 2007).Using again the additive approach, Hayes et al. (2011) found a net carbon sink fire effect on the pan-boreal carbon balance for decades of 1960s to 1990s with a similar magnitude than our study (0.03-0.08 Pg C yr −1 ).They argue that fires have changed from a carbon sink to source term for the 2000s decade (ca.−0.13 Pg C yr −1 ) due to increased fire activities (Fig. 3 in Hayes et al., 2011), which is different from our conclusion.However, it should be noted that their estimated pan-boreal carbon sink for 1997-2006 (0.04 Pg C yr −1 ) was much lower than those based on atmospheric inversion or inventory approaches (Ciais et al., 2013).On the other hand, their estimated fire-carbon emissions (0.3 Pg C yr −1 for north of 45 • N) are 50 % higher than GFED3.1 data.Thus it is likely that the biases in their estimated carbon fluxes (overestimation of emissions and underestimation of carbon sink) could lead to over-estimation of the carbon source effect by fires in the 2000s decade.Finally, Yuan et al. (2012) examined the effect of changes in fire regime on the carbon balance of the Yukon River Basin depend on how the post-fire recovery sink and fire-carbon emissions respond to the changes in climate and atmospheric CO 2 concentration

Uncertainties and future perspective
As the version of ORCHIDEE used here does not include explicit forest stand structure and successional dynamics (age classes) within grid cells, we are unable to distinguish between the ecosystem effects of surface and crown fires.Instead, simulated fire effects (e.g., fuel combustion completeness, tree mortality) are applied to the whole grid cell in proportion to the burned fraction, as is done in most other fire models (Kloster et al., 2010;Li et al., 2012;Pfeiffer et al., 2013).Due to this inability to characterize the sub-grid level fire regime, fires seldom lead to complete destruction of the whole forest stand and re-establishment of a new cohort at the grid cell level (because the burned fraction seldom approaches unity).Instead, live biomass is removed in proportion to the simulated mortality multiplied by the simulated burned fraction.As forest is never completely killed, this approach might lead to a faster post-fire recovery in the model compared with that after a crown fire in reality.Our finding that the legacy sink peaked in the decade of 1990s might be biased by this model behavior.Due to lack of explicit forest structure and vertical profile, the model is not able to simulate the thinning effects of surface fires.However, the evolution of fire impacts on the simulated NBP with time-since-disturbance on the regional scale (Fig. 6) generally resembles the temporal pattern of post-fire forest NEP observed at site level (e.g., Fig. 1 in Amiro et al., 2010), that is, a carbon source effect at the time of and for a few years after fire occurrence, followed by long-term decaying sink effect.Besides the uncertainties introduced by the model's inability to distinguish crown fire vs. surface fire, underestimation of burned area in central Asian grasslands and eastern Siberian boreal forests is another source of uncertainty in our results.We expect the underestimation of grassland burned area to make little impact on the estimated fire legacy sink effects, as grasslands quickly recover from fires, thus over a centennial time scale their fire legacy impact on NBP would be close to zero.The underestimation Introduction

Conclusions References
Tables Figures

Back Close
Full of forest fire burned area in eastern Siberia, on the other hand, might lead to an underestimation of fire legacy sink effect, as it is clear that crown fires create a long-term sink and surface fires also result in enhanced forest growth due to a short-term increase in available resources (Schulze et al., 2012).However, it is difficult to quantify the uncertainties in our results by comparing them with observational data.For one thing, as forest age is not explicitly simulated within each grid cell, no forest age map could be derived from our model simulation; this precludes evaluating our results against inventory-based forest age maps.Despite the fact that a current-day forest age map has been compiled for boreal North America (Pan et al., 2011a;Stinson et al., 2011), those for boreal Eurasia are still scarce.Further, the reconstruction of historical forest age dynamics will need a hindcast of the current forest age map by combining it with known disturbance histories.Geospatially explicit burned area data sets are available for Alaska, USA and Canada staring from 1950s (Kasischke et al., 2010;Stocks et al., 2003); those for Russia are only available starting satellite-based mapping of burned area (Giglio et al., 2013) and existing reconstructed data were based on simple assumptions and subject to great uncertainties (Balshi et al., 2007;Mouillot and Field, 2005).To derive a better estimate of the role of fire in the boreal carbon cycle requires a two pronged approach: collecting historical fire data for the Eurasian boreal region and further model developments to include forest age groups in ORCHIDEE (Naudts et al., 2015).Introduction

Conclusions References
Tables Figures

Back Close
Full Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | past fires during 1850-2009 to the current carbon balance (2000-2009) in the panboreal region (44-84 • N).The tool used is the global dynamic vegetation model OR-CHIDEE with the prognostic fire module SPITFIRE.Fire occurrences are simulated in a prognostic way, with the dynamic vegetation module being activated.Our objectives are: (1) to compare the simulated vs. observed distribution of tree cover and tree groups, with the presence of fire disturbance, (2) to separate the legacy sink of past fires from emissions of current fires to the pan-boreal carbon balance, and further quantify the relative sink contributions by fires in different decades of the past.Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | ulation protocol and allows us to separate legacy carbon sinks from past fires to the carbon balance for the 2000s decade (2000-2009) from emissions by current fires.This conceptual framework was inspired by the theoretical attribution framework on the role of land use change in carbon balance by Gasser and Ciais (2013).Evidences indicate that currently observed terrestrial carbon sink is related with the environmental perturbations in contrast with the pre-industrial time (ca.before 1850).These perturbations notably include climate change, atmospheric CO 2 concentration increase and Discussion Paper | Discussion Paper | Discussion Paper |

F
ON (g, b) is the total carbon balance of the area S(g, b) typically expressed in g C yr −1 with presence of fire, and all lowercase f functions indicate the intensive carbon balance expressed as g C m −2 yr −1 for various cases: Discussion Paper | Discussion Paper | Discussion Paper | land impacted by the CCN perturbation (thus not equal to zero), f c (g, b) is the firegenerated cohort carbon flux density without the CCN perturbation, ∆f c (g, b) is the deviation of carbon flux from a cohort under steady environment conditions because of the CCN perturbation (Fig. 1b blue curve).δS i is the fire-disturbed land cohorts within the i th decade, with i ranging from 1850s (1850-1859) to 2000s (2000-2009), ∆S(g, b) is the sum of disturbed land areas from fires of all decades since 1850.Note in Eq.
OFF,D (g, b) is the carbon balance for 2000s decade but with fires being suppressed in the D decade, with the contribution by fires of the D decade being simultaneously removed from the right hand of the equation.Thus, the contribution by fires of Discussion Paper | Discussion Paper | Discussion Paper | the D decade is the difference between F ON (g, b) and F OFF,D (g, b): conducted a reference simulation (SIM fireON ) from 1850 until 2011, accounting for climate change, atmospheric CO 2 concentration change and prognostically simulated Introduction Discussion Paper | Discussion Paper | Discussion Paper | The model was forced by the CRUNCEP climate data at 2• resolution, re-gridded from its original resolution of 0.5• .The CRUNCEP is a gridded climate data reconstructed from CRU data interpolated into NCEP temporal resolution (http://dods.extra.cea.fr/store/p529viov/cruncep/V4_1901_2012/readme.htm).The fire module needs additional input data for lightning flashes and human population density.Lightning flashes were retrieved from the High Resolution Monthly Climatology of lightning flashes by the Lightning Imaging Sensor-Optical Transient Detector (LIS/OTD) (http: //gcmd.nasa.gov/records/GCMD_lohrmc.html).The LIS/OTD dataset provides annual mean flash rates over the period of 1995-2000 at 0.5 • scale with monthly time step, which was cycled each year throughout the simulation.Annual historical population density map was retrieved from the Netherlands Environmental Assessment Agency (http://themasites.pbl.nl/tridion/en/themasites/hyde/download/index-2.html).Both lightning data sets were re-gridded at 2 • resolution before being fed into the model.The reference simulation SIM fireON consists of a spin-up run from bare soil and a transient run, with the fire module being activated.For the spin-up, climate data for the period 1901-1930 were cycled, and atmospheric CO 2 concentration (285 ppm) and pop-Discussion Paper | Discussion Paper | Discussion Paper |ulation density were prescribed at the 1850 level.The spin-up run lasted for 400 years, but contained three runs of soil-only processes each lasting 1000 years to speed up reaching equilibrium for slow and passive soil carbon pools.We verified that the average annual NBP during the last 30 years of the spin-up run was −0.003 Pg C yr −1 (a negative value as the model recovers from fast accumulation of soil carbon in the soil-only runs) and that no significant trend exists for annual NBP, indicating that the model had approximately reached an equilibrium state.The spin-up was followed by a transient simulation for 1850-2011, in which transient climate data, atmospheric CO 2 concentration and population density data were used.For 1850-1900, cycling climate data of 1901-1930 continues to be used.
Discussion Paper | Discussion Paper | Discussion Paper | panels 1a and 1b).Both MODIS land-cover data and ORCHIDEE simulation indicate Discussion Paper | Discussion Paper | Discussion Paper | simulated annual NBP for 1850-2011 for the study region in non-agricultural land and contributions of decadal fire cohorts to the carbon balance after the fire occurrence are shown in Fig. 6.The simulated annual carbon sink by the reference simulation for Discussion Paper | Discussion Paper | Discussion Paper | 1990-2011 is 0.91 Pg C yr −1 (Fig. 6a), which falls within the range of forest inventorybased estimates (∼ 0.7 Pg C yr −1 by Pan et al., 2011b) and the mean value of the terrestrial carbon cycle models (∼ 1.1 Pg C yr −1

Figure 7
Figure 7 shows the contributions of fires within each decade to the annual NBP of the study region for 2000-2009.All decades before 2000 cause a fire legacy sink, collectively having a total sink of 0.23 Pg C yr −1 .These legacy sinks are compensated by a carbon source of 0.17 Pg C yr −1 by fires within 2000-2009, leaving a net fire effect of 0.06 Pg C yr −1 .This net sink fire effect represents only a very small fraction (6.3 %) of the simulated annual carbon sink by the reference simulation (0.95 Pg C yr −1 ), indicating that most of this sink occurs in unburned natural ecosystems for which the model produces enhanced carbon storage due to climate warming (e.g., longer growing seasons) and the CO 2 fertilization effect.The sink contributions of different decadal fire cohorts (1850-1999) exhibit a general decaying trend as the cohort ages, with the variations being affected by changes in climate, atmospheric CO 2 concentration and fire disturbance.Fires in the most recent four decades (1960-1999, i.e., corresponding to a "cohort age" of 10-40 years) collectively contribute 0.14 Pg C yr −1 , accounting for 61 % of total legacy sink effect.Fires in the past century (1900-1999) contribute 0.19 Pg C yr −1 , or 83 % of the total legacy sink.
Discussion Paper | Discussion Paper | Discussion Paper | Fireswith intermediate FRIs(50-200 yr)  contribute by 0.06 Pg C yr −1 (26.1 % of the total sink effect), while vary rare fires (with an FRI > 500 yr) or very frequent fires (with an FRI of 2-10 yr) contribute least to the total sink effect (collectively contributing 0.04 Pg C yr model performance, vegetation dynamics and fire burned area ORCHIDEE-SPITFIRE successfully captured the large-scale spatial pattern of tree cover distribution, and the distribution of broadleaf vs. needleleaf and evergreen vs. deciduous forests in different continents, with the presence of fire disturbances being prognostically simulated.The simulated boreal carbon sink for the 1990s and 2000s decade is comparable with other independent approaches, with simulated fire-carbon emissions being close to GFED3.1 data.To our knowledge, this simulation of dynamic vegetation, with simultaneous constraining of simulated carbon balance and fire-carbon emissions, has been reported for the first time by using a global DGVM for the panboreal region.The larger spatial extent of deciduous needleleaf forests in Siberia and northern regions of America in ORCHIDEE might be related with our DGVM parameterization that, winter extreme coldness leads to elevated mortality of all forests except deciduous needleleaf ones; this expands their presence within the treeline limit as represented by an isotherm of growing-season soil temperature(Zhu et al., 2015).Schulze et al. (2012) found that in a transitional zone (61-64 • N, 90-107 • E) in central Siberia, where the species Picea obovata and Abies sibirica (evergreen conifers) Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | forests in Alaska from 1960 to 2006 by comparing simulations with changed and stationary fires.They found increased fires, compared with a stationary fire regime, have reduced the total ecosystem carbon storage by 185 Tg C, or 4 Tg C yr −1 .Despite not the exact same simulation approach, we also found a net carbon source fire effect of 1.5 Tg C yr −1 for the 2000s-decade carbon balance for Alaska, in the same direction as Yuan et al. (2012) but with a smaller magnitude.Our results highlight important contributions of past fire disturbances to the current ecosystem carbon sink, thanks to post-fire vegetation recovery being enhanced by CO 2 fertilization and climate warming.The latter two factors, in spite of their roles not being disentangled in the current study, might also influence the occurrence of fires and their emissions in the 2000s decade, which partially counteract the sink effects by previous fires.In the long term, change in ecosystem structure and species will also affect fuel load and combustion completeness and modify fire emissions as well.Therefore, the future role of fires in the carbon balance of boreal regions remains rather uncertain and Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper |

Figure 1 .Figure 3 .Figure 4 .Figure 5 .Figure 7 .Figure 8 .
Figure 1.A conceptual framework to attribute current-day boreal carbon sink.(a) A schematic graph showing how the carbon balance of a geographical point (with a total area of S) for the 2000s decade is composed of carbon fluxes from undisturbed mature forests, forest cohorts as legacies of past decadal fires, and fire-carbon emissions within the 2000s decade.Indicative examples are given for cohorts generated by fires of the 2000s, 1990s, 1940s and 1850s decades.(b) The evolution of forest net ecosystem productivity (NEP) with the time-sincedisturbance after fire under pre-industrial conditions and as impacted by the CCN perturbation.