The role of historical fire disturbance in the carbon dynamics of the pan-boreal region: A process-based analysis

Authors


Abstract

[1] Wildfire is a common occurrence in ecosystems of northern high latitudes, and changes in the fire regime of this region have consequences for carbon feedbacks to the climate system. To improve our understanding of how wildfire influences carbon dynamics of this region, we used the process-based Terrestrial Ecosystem Model to simulate fire emissions and changes in carbon storage north of 45°N from the start of spatially explicit historically recorded fire records in the twentieth century through 2002, and evaluated the role of fire in the carbon dynamics of the region within the context of ecosystem responses to changes in atmospheric CO2 concentration and climate. Our analysis indicates that fire plays an important role in interannual and decadal scale variation of source/sink relationships of northern terrestrial ecosystems and also suggests that atmospheric CO2 may be important to consider in addition to changes in climate and fire disturbance. There are substantial uncertainties in the effects of fire on carbon storage in our simulations. These uncertainties are associated with sparse fire data for northern Eurasia, uncertainty in estimating carbon consumption, and difficulty in verifying assumptions about the representation of fires that occurred prior to the start of the historical fire record. To improve the ability to better predict how fire will influence carbon storage of this region in the future, new analyses of the retrospective role of fire in the carbon dynamics of northern high latitudes should address these uncertainties.

1. Introduction

[2] Terrestrial ecosystems in high latitudes contain large reserves of carbon [McGuire et al., 2002, 2006]. Wildfire is a common disturbance that affects the structure and function of ecosystems in the region [McGuire et al., 2006]. Pronounced warming in high latitudes, which has been occurring for the past several decades [Chapman and Walsh, 1993; Serreze et al., 2000; Serreze and Francis, 2006; McGuire et al., 2006], is altering the fire regime of the region [Gillett et al., 2004; Kasischke and Turetsky, 2006] and has consequences for carbon storage of northern ecosystems [Kasischke et al., 1995; Stocks et al., 1998; Flannigan et al., 2005]. While many studies have focused on using fire observation data to estimate fire emissions in northern high latitudes [Conard and Ivanova, 1997; French et al., 2000; Shvidenko and Nilsson, 2000; Kajii et al., 2002; Conard et al., 2002; Kasischke and Bruhwiler, 2002; Potter et al., 2003a; Soja et al., 2004; Yurganov et al., 2004; Kasischke et al., 2005], understanding the role of fire on carbon dynamics in this region requires consideration of several additional factors.

[3] The state of the landscape, or stand age distribution across the landscape before a fire event occurs is one of the factors that influences carbon dynamics [Kurz and Apps, 1999; Chen et al., 2002, 2003]. Stand age distributions in fire-prone systems are directly affected by the historical patterns of fire across the landscape. Although data sets exist that provide an historical picture of fires across the landscape, estimating the effects of fire on carbon dynamics requires that fires are accounted for prior to the start of the historical record [McGuire et al., 2004]. While several studies have used fire cycle information to account for recurring fires, they generally do not explicitly consider the history of fire across the landscape [e.g., Thonicke et al., 2001; Venevsky et al., 2002].

[4] Another important factor to consider when estimating the effects of fire on carbon dynamics is influence of burn severity, which can be defined as the fractional amount of carbon consumed during a fire from both aboveground and ground layer biomass [Kasischke et al., 2005]. Burn severity is highly variable in northern ecosystems, and depends on the time of the year in which the fire occurs [Kasischke et al., 1995, 2000], amount of fuel [Furyaev, 1996], spatial heterogeneity of vegetation and topography across the landscape [Turner and Romme, 1994], and weather conditions at the time of fire [Johnson, 1992]. As a result, representing burn severity across large spatial scales has proven to be difficult and is typically associated with a particular vegetation type or ecoregion [French et al., 2002]. Furthermore, the amount of carbon consumed on a per-fire basis can differ with respect to the type of fire regime, which is defined by the intensity, frequency, seasonality, size, and type of fire [Weber and Flannigan, 1997]. In the North American boreal forest fires are predominantly stand-replacing and characterized by a high-intensity crown fire regime [Johnson, 1992]. Fires that occur across boreal Eurasia range from low-intensity surface fires (e.g., Siberian Scots pine stands) to high-intensity crown fires that dominate boreal needle-leaf, larch (deciduous conifer), and pine stands (evergreen conifer) [Conard and Ivanova, 1997; Wirth et al., 2002a].

[5] To understand the role of fire in the carbon dynamics of northern ecosystems, it is also important to evaluate changes in fire disturbance in the context of other environmental changes. While several studies have been conducted that incorporate the influence of fire on carbon dynamics in the context of forest inventory data [Kurz and Apps, 1999; Shvidenko and Nilsson, 2002, 2003; see also Myneni et al., 2001], these studies do not explicitly consider the effects of other environmental factors such as changes in atmospheric CO2 and climate. Process-based models are designed to evaluate how changes in climate and environmental chemistry influence carbon dynamics, and simulations can be conducted to quantify the effect of individual factors [McGuire et al., 2001]. Process-based models also complement estimates of regional carbon storage made by atmospheric inversion models [Schimel et al., 2001; Dargaville et al., 2002, 2006; Gurney et al., 2004], which collectively can identify uncertainties in the net exchange between the earth's surface and the atmosphere, but are not able to evaluate the mechanisms responsible for the exchanges.

[6] Several process-based studies have been conducted that incorporate the influence of disturbance on carbon dynamics but focus primarily on its response to land-use change [McGuire et al., 2001] or regional fire regimes [Peng and Apps, 1999; Amiro et al., 2000; Venevsky et al., 2002; Chen et al., 2000, 2003]. Other process-based studies have used the satellite record to infer disturbance, but have not explicitly considered the role of fire dynamics prior to the start of the satellite record [e.g., Potter et al., 2003a, 2003b, 2005]. Mouillot et al. [2006] used a process-based model to estimate fire emissions, but do not estimate the overall effect of fire on the carbon budget. To our knowledge, a study conducted by Zhuang et al. [2006] is the only analysis that uses a process-based approach to simulate the effects of fire on northern ecosystems using historical fire records. However, Zhuang et al. [2006] did not consider how carbon dynamics are influenced by spatial variability in burn severity and spatial variability in fire frequency prior to the start of the historical record. The observed changes in climate [Chapman and Walsh, 1993; Serreze et al., 2000] and the potential for a changing climate to alter future fire regimes of northern high latitudes [Wotton and Flannigan, 1993; Flannigan et al., 1998; Kasischke et al., 1995; Stocks et al., 1998; Wotton et al., 2003; Flannigan et al., 2005; McCoy and Burn, 2005] suggest that it is important to project how future changes in carbon dynamics respond to changes in the fire regime. Our ability to make projections of future changes in carbon dynamics of northern ecosystems is limited by our understanding of how the temporal and spatial aspects of fire influence historical carbon dynamics.

[7] The focus of this study is to improve our understanding of the role of historically recorded fire on carbon dynamics in ecosystems of northern high latitudes north of 45°N (referred to hereafter as the “pan-boreal region”). In particular, our objectives are to estimate fire emissions and changes in carbon storage in the pan-boreal region, to evaluate the role of historically recorded fire in carbon dynamics of the region in the context of ecosystem responses to changes in atmospheric CO2 concentrations and climate, and to identify sources of uncertainty that should be reduced in retrospective analyses of the role of fire in the carbon dynamics of the pan-boreal region. In comparison to a previous study by Zhuang et al. [2006], our analysis considers how carbon dynamics are influenced by spatial variability in burn severity and by spatial variability in fire frequency prior to the start of the historical record of fire in terrestrial ecosystems of northern high latitudes. We also identify key sources of uncertainty that should be reduced in order to better understand the role of fire in the carbon dynamics of the pan-boreal region.

2. Methods

2.1. Overview

[8] In this study we evaluate how changes in atmospheric CO2 concentration, climate, and fire influence carbon dynamics for North America and Eurasia north of 45°N using the process-based Terrestrial Ecosystem Model (TEM). The advantage of using a process-based model for simulating carbon dynamics is that individual processes that affect carbon storage can be isolated. To initialize our simulations we first run the model to equilibrium (annual net primary production = annual heterotrophic respiration) in year 1000 for each terrestrial 0.5° (latitude by longitude) grid cell north of 45°N using the mean monthly climate from 1901–1930. We then conduct a 900 year spin-up (from year 1001–1900) to dynamically equilibrate the model to the fire regime and to multidecadal variability in the climate. During the spin-up period, climate for the period 1901–1930 is repeated. A backcasting approach (see section 2.4) is used to account for the influence of fire on carbon dynamics (including the spin-up period) before 1959 for North America and before 1996 for Eurasia. The model is then run from year 1901–2002 using gridded monthly climate based on observations (see section 2.3). In this study, we conduct two sets of simulations. In the first set of simulations, photosynthesis is sensitive to increasing atmospheric CO2 concentrations (a CO2 fertilization effect), while in the second set photosynthesis is not sensitive to increasing atmospheric CO2 concentrations. For the set considering the effect of atmospheric CO2 fertilization we conduct three simulations. In simulation one (S1), atmospheric CO2 concentration is allowed to vary, but a mean monthly climate from 1901–1930 is used to represent climate for each year (i.e., “constant climate”) and no fire disturbances are assumed to occur. In simulation two (S2), both atmospheric CO2 concentration and climate are allowed to vary, but again, no fire disturbances are assumed to occur. In simulation three (S3), atmospheric CO2 concentration and climate are allowed to vary and fire disturbances are assumed to occur. For the second set of simulations we conduct the same three simulations as described in the first set, but with atmospheric CO2 fixed at 296 ppm, which is the mole fraction used to initialize each simulation. We then analyze our simulation results for the periods of historically recorded fire disturbance, which are 1959–2002 in boreal North America and 1996–2002 in the pan-boreal region. The effect of CO2 fertilization on carbon storage is determined by the results of the S1 simulation. The effect of climate on carbon storage is determined as the difference in results between the S2 and S1 simulations. Similarly, the effect of fire on carbon storage is determined as the difference in results between the S3 and S2 simulations.

2.2. Terrestrial Ecosystem Model (TEM)

[9] The TEM is a large-scale, process-based biogeochemical model that estimates monthly pools and fluxes of carbon and nitrogen for land-based areas. The model is coupled to a soil thermal model and can be applied on both permafrost and nonpermafrost soils [Zhuang et al., 2003]. The TEM is driven by a series of spatially explicit data sets that include climate, elevation, soil texture, and vegetation. The equations and parameters of TEM have been documented in previous studies [Raich et al., 1991; McGuire et al., 1992; Tian et al., 1999; Zhuang et al., 2003; Euskirchen et al., 2006]. The model has been applied previously to various regions across the globe including northern ecosystems [e.g., McGuire et al., 2000a, 2000b, 2001, 2002, 2004; Clein et al., 2000, 2002; Zhuang et al., 2001, 2002, 2003; Euskirchen et al., 2006]. Our application of TEM to this study is based on version 5.1 of the model [see Euskirchen et al., 2006], which has been modified in this study to incorporate the effects of fire (Figure 1). Several of the parameters within TEM are defined based on values obtained from the peer-reviewed literature. However, the rate limiting parameters are defined by calibrating the model to pools and fluxes of field sites representative of particular ecosystems (e.g., tundra and boreal forest). To estimate changes in carbon storage, we calculated the Net Ecosystem Carbon Balance (NECB [see Chapin et al., 2006]) for outputs generated by the model as:

equation image

where NPP is net primary production, Rh is heterotrophic respiration, and TCE is total carbon emitted due to fire. It is important to note that our analysis does not consider the effects of other disturbances that affect carbon storage in the pan-boreal region, for example, insect disturbance, forest harvest, or land-use change, in the calculation of NECB.

Figure 1.

The simulation framework of this study in which the Terrestrial Ecosystem Model (TEM) was used to simulate the effects of fire on carbon dynamics. Input data sets include CO2 concentration, cloudiness, air temperature, precipitation, and spatially explicit information on fire history (area burned), burn severity (carbon fraction consumed during a fire event), and fire return interval (FRI, used for inserting fires prior to the historical record). Burn severity parameters, fire history, and FRI are used to calculate fire emissions from TEM carbon pools (vegetation and soil carbon). Fire regime also has indirect effects on net primary production (NPP) and heterotrophic respiration (RH) through the influence on soil and vegetation carbon pools. Model outputs are NPP, RH, and fire emissions which are used to calculate the net ecosystem carbon balance (NECB).

2.3. Input Data Sets

[10] To extrapolate TEM across North America and the pan-boreal region, we used driving data sets that had (1) temporal variability, but no spatial variability (atmospheric CO2 concentration), (2) spatial variability but no temporal variability (elevation, soil texture, and vegetation), and both temporal and spatial variability (air temperature, precipitation, cloudiness, and fire disturbance). Below, we describe these data sets in more detail.

2.3.1. Atmospheric CO2, Elevation, Soil Texture, and Vegetation Data Sets

[11] In this study, atmospheric CO2 data were obtained from the Mauna Loa station [Keeling and Whorf, 2005]. TerrainBase v1.1 elevation data were obtained from the National Geophysical Data Center, Boulder, CO [NGDC, 1994] and aggregated to a 0.5° latitude × 0.5° longitude spatial resolution. Soil texture, represented as percent silt plus percent clay in TEM, was based on the Global Gridded Surfaces of Selected Soil Characteristics data set [Global Soil Data Task Group, 2000] and gridded at a 0.5° latitude × 0.5° longitude spatial resolution. The input vegetation data set, gridded at the 0.5° resolution, is represented by a potential natural vegetation map described by Melillo et al. [1993].

2.3.2. Temperature, Precipitation, and Cloudiness Data Sets

[12] Monthly air temperature (°C), precipitation (mm), and cloudiness (%) data derived from observations for the period 1901–2002 gridded at 0.5° resolution were obtained from the Climate Research Unit, University of East Anglia [Mitchell and Jones, 2005].

2.3.3. Historical Fire Data Sets

[13] A database of fire point location data and 1-km resolution fire scar data sets were acquired for Alaska, Canada, and Eurasia and then assembled into a 0.5° grid. For Alaska, we used the Alaska fire scar location database initially developed by Kasischke et al. [2002] and maintained by the Bureau of Land Management [2005]. The database contains point and boundary location information for fires in Alaska from 1950–2002. Fires greater than 1000 acres (∼404 ha) are included from 1950–1987, inclusive, and fires greater than 100 acres (∼40.4 ha) are included from 1988–2002, inclusive. Although our analysis is focused on the region north of 45°N, fires in the northern conterminous United States are not considered.

[14] For Canada we used a combination of point location data from the Canadian Large Fire Database (LFDB) and provincial polygon data, with a preference for using the provincial polygon data when available. The LFDB is a compilation of provincial and territorial wildfire data that represents all fires that are greater than 200 ha that occurred from 1959–1999. For the point location data sets for Canada [Flannigan and Little, 2002], we used the longitudinal and latitudinal point locations to calculate a radius for each location based on the area of the historical fire. Circular fire boundaries were then created for each point by buffering each point by a distance equal to the calculated radius. The provincial polygon data represent fires in all provinces from 1980–2002 (provided by M. Flannigan, Canadian Large Fire Database, 1980–2003 polygons, unpublished data, 2006). Also, historical fire data for Saskatchewan [Naelapea and Nickeson, 1998] and Alberta [Government of Alberta, 2005] were also available as polygon coverages for the periods 1945–1979 and 1931–1979, respectively. There was no redundancy between the use of point location data of the Canadian LFDB and the provincial polygon data in our assembly of the historical data set of fire in Canada for use in our simulations.

[15] For Russia, we used Advanced Very High Resolution Radiometer (AVHRR) satellite-derived fire scars data from 1996–2002 produced at the Sukachev Institute of Forestry in Krasnoyarsk [Sukhinin et al., 2004].

[16] Our examination of the spatially explicit fire scar data indicated that there were a number of spatial units within each 0.5° grid cell that had unique fire histories over the length of the fire scar record. These unique fire histories result in stands of different age that have different legacies of fire disturbance on carbon storage within a 0.5° grid cell. To properly represent this legacy of disturbance within a 0.5° grid cell, we labeled each spatial unit within a 0.5° grid cell that has a unique fire history based on the fire scar record as a “cohort”. The number of cohorts per grid cell depended on both the historical fire record and fires that we inserted prior to the start of the historical fire record as part of backcasting algorithm (see section 2.4). To estimate carbon storage changes for a 0.5° grid cell, we conducted simulations for each cohort within the grid cell and aggregated the simulated carbon storage estimates across all of the cohorts of the grid cell.

2.4. Fire Return Intervals and Backcasting

[17] To take into account fires prior to the start of the historical fire record, we developed a backcasting algorithm which requires information on the fire return interval (FRI). We defined FRI as the time required to burn an area equal to the entire 0.5° grid cell. Each cohort within a given 0.5° grid cell has the same FRI regardless of when the cohort burned historically. For North America, we calculated FRI based on the historical fire record from 1950–2002 in Alaska and 1959–2002 in Canada. This was accomplished by taking into account the proportion of a grid cell burned each year by first calculating a fire rate (FR) given by:

equation image

in which AB is the area burned within a 0.5° grid cell, AT is the total area of the 0.5° grid cell, and NY is the number of years representing the historical fire record. Since FRI is the time required to burn an area equal to the entire 0.5° grid cell, it is calculated as the inverse of the fire rate:

equation image

The FRI map as calculated above was then smoothed using a nearest-neighbor method in order to be more spatially representative of fire regime by reducing pixilation (Figure 2a).

Figure 2.

Fire return interval (FRI) maps for (a) North America and (b) Eurasia. North America FRIs were based on the proportion of a 0.5 degree cell burned over the historical fire record (1950–2002 for Alaska; 1959–2002 for Canada). Eurasian FRIs were interpolated using ordinary cokriging methods based on nontemporally explicit literature estimates.

[18] A different approach was used for estimating FRI for Eurasia (Figure 2b) because of the short length of the historical fire record as well as the lack of large-scale FRI data. FRIs were estimated based on available data using ordinary cokriging methods in the ESRI ArcMap v9.0 Geostatistical Analyst Extension Package. The available FRI data for Eurasia were obtained in the form of nontemporally explicit points provided by C. Wirth (unpublished data, 2006) and transects. Transect data are based on the IGBP high-latitude transect study of McGuire et al. [2002]. Vegetation data at 1-km resolution (E. S. Euskirchen et al., Energy feedbacks to the climate system associated with snow cover dynamics in northern high latitudes during warming periods of the 20th century, submitted to Global Change Biology, 2007) were used as a second predictor variable to help improve the interpolated surface. Because the fire scar record for boreal Eurasia is so short, we then adjusted the initial Eurasia FRI estimates based on the assumption that the ratio of mean annual area burned from 1996–2002 to long-term mean annual area burned was similar over the long-term in boreal Eurasia and Canada. To implement this assumption, the interpolated surface of the initial FRI (IFRI) estimates was standardized relative to a factor δ calculated from historical burn area for 1996–2002 and interpolated FRIs in Eurasia and Canada as:

equation image

in which δ is calculated as:

equation image

in which

equation image

and

equation image

where ϕC and ϕE are the respective burn ratios for Canada or Eurasia, μC and μE are the respective mean annual areas burned from 1996–2002 for Canada and Eurasia, and μFRICanada and μIFRIEurasia are the respective mean annual areas burned based on interpolated fire return intervals for the boreal forest area of Canada and on the initial interpolated fire return intervals for the boreal forest area of Eurasia.

[19] Throughout the pan-boreal region, the interpolated FRIs were then used by the backcasting algorithm to insert fires prior to the start of the historical period based on the fire record of each cohort within a 0.5° grid cell and the FRI of that grid cell. Fires were inserted by one of two ways. If a given cohort burned over the length of the historical period, previous fire(s) events were calculated by the difference between the first historical burn year and the FRI. If the cohort did not burn during the historical fire record, fires were inserted stochastically based on the FRI of the grid cell prior to the historical fire record. Backcasting fires only occurred if the grid cell FRI was less than or equal to 500 years (i.e., each cohort would burn at least two times during a 900 year spin-up period), allowing a dynamic equilibrium to be reached prior to the start of the realistic transient climate period (1901–2002). Fires were not inserted for Europe (defined as west of 22°E and north of 45°N in this study) because we assumed that human activities have effectively suppressed wildfire in this region; the historical fire record we used for Russia did not contain any fires west of 22°E and north of 45°N.

2.5. Burn Severity Implementation

[20] Our approach to modeling emissions is based on calculating the total carbon emitted during a fire event from aboveground and ground layer carbon consumption estimates. Literature estimates (Table 1) of aboveground and ground layer carbon fraction consumed during a fire for boreal North America [French et al., 2000] and boreal Eurasia [FIRESCAN Science Team, 1996; Kajii et al., 2002; Wirth et al., 2002b] are used to address the issue of burn severity. Total annual carbon emissions are then calculated using these parameters by calculating fluxes for both vegetation and soil carbon pools during a fire by:

equation image

where TCE is the total carbon emitted, βa is the aboveground C fraction consumed, βg is the ground layer carbon fraction consumed during a fire, Vc is vegetation carbon, and Sc is soil carbon. Based on Harden et al. [2004] and Wirth et al. [2002a], we assumed that 85% of soil and vegetation nitrogen was retained at the time of fire. The nitrogen lost from the ecosystem as a result of fire was reintroduced into the system annually in equal increments obtained by dividing the total net nitrogen lost to the atmosphere during the most recent fire event by FRI.

Table 1. Literature Estimates of Average Aboveground (βa) and Ground Layer (βb) Carbon Fraction Consumed Used for Emissions Estimates During a Fire Event for North America [French et al., 2000] and Eurasia [FIRESCAN Science Team, 1996; Kajii et al., 2002; Wirth et al., 2002a, 2002b]a
EcozoneAboveground (βa) C Fraction ConsumedGround Layer (βb) C Fraction ConsumedAverage Area Burned, haAverage Emission, Tg C yr−1Average Emission per m2 of Burned Area, g C m2 yr−1
  • a

    Also shown are mean annual area burned, mean annual total carbon emission, and mean annual total carbon emission per square meter of burned area from model simulations for North America (1959–2002) and Eurasia (1996–2002).

North America
Alaska Boreal Interior0.230.362890007.22470
Boreal Cordillera0.130.381590005.73580
Taiga Plain0.250.063620006.01650
West Taiga Shield0.250.053690003.3896
East Taiga Shield0.250.051410002.11490
West Boreal Shield0.260.0653100015.22860
East Boreal Shield0.220.06950000.2256
Boreal Plain0.240.112270007.83420
Hudson Plain0.240.05563000.81430
Eurasia
Larch Forests0.150.282090000106.85110
Ground fire regime0.150.15254000073.32880
Grassland/Steppe0.850.0175300035.64720

[21] We also differentiated between crown and surface fires in our simulations. For boreal North America we assumed a fire regime that was predominantly stand replacing and specified that one percent of live plant biomass would be available for regeneration following a fire. For Eurasia, we assumed a stand-replacing fire regime for larch forests across eastern Siberia and grassland/steppe at the southern boundary of our study region. Areas east of 22°E not dominated by larch forests or grassland were classified as being driven by a surface fire regime, and we assumed that 60% of aboveground vegetation remains following fire events [Wirth, 2005].

3. Results

[22] We first present our estimates of fire emissions across North America and the pan-boreal region. We then examine the relative importance of these fire emissions to other environmental factors in the carbon dynamics of terrestrial ecosystems in North America and the pan-boreal region. North America is highlighted because we had a longer period of historical fire data for this region (1959–2002) than for the entire pan-boreal region (1996–2002).

3.1. Fire Emissions

[23] Fire emissions calculated by TEM are presented as total carbon lost to the atmosphere at the time of a fire event. We calculated decadal averages to examine the long-term trends in simulated fire emissions for boreal North America. The results of our simulations indicate that the decadal average annual fire emissions for Alaska, Canada, and North America (Alaska and Canada combined) approximately doubled from the 1960s to the 1980s and that CO2 fertilization had little effect on the estimated emissions (Figure 3a). Although a slight decrease in average fire emissions from the 1980s to 1990s was simulated for Canada (and boreal North America), simulated fire emissions for Alaska nearly doubled.

Figure 3.

Fire emissions of total carbon: (a) average decadal emissions for Alaska and Canada; and (b) annual emissions for Eurasia and North America. Units are Tg C yr−1.

[24] In our pan-boreal simulations from 1996–2002, boreal Eurasia accounted for approximately 80% of estimated emissions and CO2 fertilization had little effect on emissions estimates. Across the pan-boreal region the estimated mean annual emissions of total carbon from 1996–2002 as a result of wildfire were 262.5 Tg C yr−1 and 254.5 Tg C yr−1 for the simulations that considered and excluded the effect of atmospheric CO2 fertilization, respectively (Figure 3b). For Eurasia mean annual emissions of total carbon were 215.7 Tg C yr−1 and 208.8 Tg C yr−1 for the simulations that considered and excluded atmospheric CO2 fertilization, respectively. The mean annual emissions of total carbon for the simulations that considered and excluded CO2 fertilization for North America were 46.8 Tg C yr−1 and 45.7 Tg C yr−1, respectively. For the North American sub-regions of Alaska and Canada, mean annual total carbon emissions for the simulation that considered CO2 fertilization were estimated to be 13.9 Tg C yr−1 and 32.9 Tg C yr−1, respectively, while the simulation that excluded CO2 fertilization estimated emissions to be 13.7 Tg C yr−1 and 32.2 Tg C yr−1.

[25] To understand the spatial variability of emissions among subregions with different burn severity parameters (Table 1), we calculated the mean annual area burned, mean total annual carbon emissions, and mean annual total carbon emissions per square meter of burned area for subregions of North America and Eurasia for the periods 1959–2002 and 1996–2002, respectively (Table 1). Across North America, the mean emissions per unit area burned was greatest across the Boreal Cordillera, Boreal Plain, West Boreal Shield, and the Alaska Boreal Interior subregions (Table 1). In our simulations, the three highest values for ground layer fraction consumed occur in the Boreal Cordillera, the Alaska Boreal Interior, and the Boreal Plain subregions, while the highest value of aboveground fraction consumed occurs in the West Boreal Shield (Table 1). Among the three subregions in Eurasia, the stand-replacing regime of the larch forest subregion, which has the highest value of ground layer fraction consumed in Eurasia (Table 1), was responsible for the highest carbon emissions per square meter of burned area (Table 1).

3.2. North American Carbon Dynamics 1959–2002

[26] Our simulations that considered atmospheric CO2 fertilization estimate that boreal North America was a carbon sink of 81.7 Tg C yr−1 (7.5 g C m−2 yr−1) from 1959–2002, while the simulations that excluded CO2 fertilization estimate a sink of 18.7 Tg C yr−1 (1.5 g C m−2 yr−1) over the same period (Table 2). For the case of CO2 fertilization, climate variability and CO2 fertilization were about equally responsible for sequestering carbon at a rate of 46.9 Tg C yr−1 (3.7 g C m−2 yr−1) and 50.4 Tg C yr−1 (4.0 g C m−2 yr−1), respectively, whereas fire was responsible for carbon release to the atmosphere at a rate of 15.6 Tg C yr−1 (1.2 g C m−2 yr−1). The effect of CO2 on carbon storage (Figure 4a) is generally positive across North America while the effect of climate on carbon storage shows both uptake from and release to the atmosphere (Figure 4b); release of carbon is most evident in the Canadian Archipelago, with greater release of carbon from the simulations that excluded CO2 fertilization. In regions where fires are concentrated over the period 1959–2002 (interior Alaska extending southeast from the Yukon Territory through central Canada to portions of eastern Quebec), carbon losses are observed in response to fire, with greater losses observed for the simulations that excluded CO2 fertilization, while areas not burned during this period generally responded as a carbon sink (Figure 4c). Overall, North America acts as a carbon sink in response to the combined effects of CO2, climate, and fire (Figure 4d), except for regions where fires occurred and in the Canadian Archipelago which lost carbon in response to climatic variability.

Figure 4.

Simulated mean annual net ecosystem carbon balance (NECB) of North America from 1959–2002 in response to (a) CO2 fertilization, (b) climate, (c) fire, and (d) CO2, climate, and fire. Results are presented for simulations conducted with and without a CO2 fertilization effect on photosynthesis. A control corresponding to Figure 4a for the simulations without CO2 fertilization is not presented because NECB would be zero throughout the region. Units are in g C m−2 yr−1. Positive values represent carbon sequestered by terrestrial ecosystems, while negative values represent release of carbon to the atmosphere.

Table 2. Mean Annual Changes in Carbon Storage Simulated for North America From 1959 to 2002 and for the Pan-Boreal Region From 1996 to 2002a
PeriodRegionEffects
CO2ClimateFireTotal
  • a

    Units are given in Tg C yr−1. Positive values represent carbon sequestered by terrestrial ecosystems, while negative values represent release of carbon.

With CO2Fertilization
1959–2002North America50.446.9−15.681.7
1996–2002Pan-boreal284.6136.9−15.9405.6
 Eurasia207.750.921.5280.2
 North America76.986.0−37.5125.4
 
Without CO2Fertilization
1959–2002North America−0.336.4−17.418.7
1996–2002Pan-boreal−0.236.9−41.6−4.9
 Eurasia0.1−29.4−0.1−29.4
 North America−0.366.3−41.524.5

[27] We further analyzed the effects of CO2, climate, and fire for North America in order to understand how each effect influences decadal-scale carbon dynamics (Figure 5). Our analysis indicates that increasing CO2 concentrations enhanced carbon storage per decade from the 1960s through the 1990s (Figure 5a). Similarly, carbon storage increased in response to increasing mean annual air temperature from the 1960s to the 1990s for both sets of simulations (Figure 5b). The effect of fire on carbon storage shows that the 1960s and 1970s were periods of sink activity, but that the sink weakened in the 1970s as area burned increased (Figure 5c). In the 1980s and 1990s, the effect of fire acted to release carbon to the atmosphere, with the effect being larger in the 1990s even though fire emissions were higher in the 1980s (Figure 3a). It is important to recognize that the effect of fire during a particular decade in Figure 5c is not simply correlated with fire emissions as it integrates the legacy of how fire history influences the balance between NPP and Rh on regrowing stands during the decade in addition to fire emissions during the decade. Thus, from the 1970s through the 1990s, our simulations indicate that the increase in mean annual area burned promoted decreases in carbon storage. The combined effects of CO2, climate, and fire in our simulations indicate, however, that North America acted as a carbon sink in each decade from the 1960s to 1990s (Figure 5d). The simulated sink activity generally increased over time with a slight dip in the 1980s and was greatest in the 1990s. The combined effects of climate and fire for the simulations that excluded CO2 fertilization show sink activity from the 1970s through 1990s, with an increase in sink activity from the 1960s to 1970s followed by a decrease from the 1970s to the 1980s and 1990s due to an increase in the area burned between the two decades.

Figure 5.

Decadal effects of (a) CO2, (b) climate, (c) fire, and (d) the combination of CO2, climate and fire on simulated net ecosystem carbon balance for North America from the 1960s through the 1990s. Effects are compared to model driving data of mean decadal CO2, air temperature, and area burned. Positive values represent carbon sequestered by terrestrial ecosystems, while negative values represent release of carbon to the atmosphere.

3.3. Pan-Boreal Carbon Dynamics for 1996–2002

[28] For the period from 1996 through 2002, we estimate that carbon storage of the pan-boreal region north of 45°N increased by 405.6 Tg C yr−1 (10.6 g C m−2 yr−1) in response to CO2, climate, and fire (Table 2). We estimate that about twice as much carbon has been sequestered in Eurasia than in North America. For the pan-boreal region, our simulations that considered CO2 fertilization indicate that CO2 fertilization sequestered over twice as much carbon (284.6 Tg C yr−1 or 7.5 g C m−2 yr−1) as climate variability (136.9 Tg C yr−1 or 3.6 g C m−2 yr−1), and that fire was responsible for releasing 15.9 Tg C yr−1 (0.4 g C m−2 yr−1) to the atmosphere. For both North America and Eurasia, the simulated effects of atmospheric CO2 and climate variation were responsible for sequestering carbon while fire acted to release carbon to the atmosphere. Similar to our longer-term analysis for boreal North America, the effects of CO2 and climate are similar in promoting carbon storage in boreal North America from 1996–2002. In contrast, the effects of increasing CO2 are about four times larger than the effects of climate in promoting carbon storage in Eurasia. Our simulations indicate that the effects of fire in North America are about four times larger than in Eurasia in promoting carbon release between 1996 and 2002.

[29] The simulations that excluded CO2 fertilization estimate that the combined effects of climate and fire were responsible for a release of 4.9 Tg C yr−1 (0.1 g C m−2 yr−1) to the atmosphere over the period 1996–2002 (Table 2). Of these effects, climate was responsible for sequestering 36.9 Tg C yr−1 (1.0 g C m−2 yr−1) while fire was responsible for releasing 41.6 Tg C yr−1 (1.1 g C m−2 yr−1) to the atmosphere.

[30] To better understand how CO2 fertilization, climate and fire may have influenced carbon storage in the pan-boreal region, we first analyzed the patterns of interannual variability in terrestrial carbon storage or loss. Increasing atmospheric CO2 concentration increased carbon storage from 1996–2002 (Figure 6a). Our analysis of the effect of climate on carbon storage did not identify a relationship with mean annual air temperature from 1996–2002 (Figure 6b). In comparison to the simulations that considered CO2 fertilization, the effect of climate on carbon storage in the simulations that excluded CO2 fertilization was to generally act as either a smaller sink or a greater source (Figure 6b). We evaluated relationships between the climate effect on carbon storage and associated air temperature and precipitation for each subregion (boreal Eurasia and North America) and for each vegetation type within a subregion, but at these scales we could not explain how climate variability influenced interannual variation in carbon storage with simple empirical relationships. The effect of fire on carbon storage shows that as total area burned increases, carbon storage decreases (Figure 6c). For both sets of simulations, larger fire years promoted less carbon storage than more moderate fire years. Overall, our simulations of the combined effects of CO2, climate, and fire indicate that the pan-boreal region acted as a carbon sink from 1996–2002 except for an estimated release of carbon in 2002 (Figure 6d). In contrast, the combined effects of climate and fire for simulations excluding CO2 indicate that the pan-boreal region acts as a carbon source in larger fire years.

Figure 6.

Effects of (a) CO2, (b) climate, (c) fire, and (d) the combination of CO2, climate and fire on simulated annual net ecosystem carbon balance for the pan-boreal region from 1996–2002. Effects are compared to model driving variables of annual CO2, air temperature, and total area burned. Positive values represent carbon sequestered by terrestrial ecosystems, while negative values represent release of carbon to the atmosphere.

[31] We further explored how the influence of these environmental factors on carbon storage varied spatially (Figure 7). Across the pan-boreal region increasing atmospheric CO2 promoted carbon storage (Figure 7a), while climate variability promoted both source and sink activity (Figure 7b). Across Eurasia, losses associated with climate are observed south of the Scandinavian region, the Kazakh Uplands, and the Mongolian Plateau, while in North America losses are observed in the Queen Elizabeth Islands and portions of Alberta and Saskatchewan. In Eurasia, carbon losses appear to be greater south of the Scandinavian region for the simulations without CO2 fertilization. Carbon gains associated with climate occur across Eurasia from western Europe to the Russian Far East and across North America from Alaska to Labrador. The effect of fire generally promoted losses of carbon to the atmosphere in areas that were identified as burned in the historical fire records that we used to drive our simulations (Figure 7c). The combined effects of CO2, climate, and fire generally promoted carbon storage across the pan-boreal region except for carbon losses in areas where fire occurred between 1996 and 2002 (Figure 7d). The combined effects of climate and fire also show a similar pattern for the simulations without CO2 fertilization; however, regions across Eurasia (south of the Scandinavian region to the Russian Far East) and North America (Canadian Archipelago) show greater carbon losses.

Figure 7.

Simulated mean net ecosystem carbon balance (NECB) of the pan-boreal region from 1996–2002 in response to (a) CO2 fertilization, (b) climate, (c) fire, and (d) the combination of CO2, climate, and fire. Results are presented for simulations conducted with and without a CO2 fertilization effect on photosynthesis. A control corresponding to Figure 7a for the simulations without CO2 fertilization is not presented because NECB would be zero throughout the region. Units are in g C m−2 yr−1. Positive values represent carbon sequestered by terrestrial ecosystems, while negative values represent release of carbon to the atmosphere.

4. Discussion

[32] The results presented here attempt to evaluate the historical effects of fire disturbance on carbon dynamics across the entire pan-boreal region in the context of changes in atmospheric CO2 and climate. We also discuss uncertainties with respect to the role of atmospheric CO2 fertilization in calculating the overall carbon budget. Given the spatial and temporal scales of our analysis, it is difficult to conduct a direct validation of our results. We are able, however, to compare our results with the existing regional estimates of fire emissions and carbon balance to evaluate interannual and decadal variation in our simulations.

4.1. Comparison of Fire Emission Estimates

[33] Our estimates of fire emissions reported for each set of simulations do not appear to be greatly influenced by our implementation of the effects of CO2 fertilization on photosynthesis. The simulations that considered atmospheric CO2 fertilization are between 1–14 Tg C higher than those that excluded the effect of CO2 fertilization. A number of studies have been conducted that use long-term historical fire data sets to estimate fire emissions within our study region (Table 3). For boreal North America, our estimates are 15–31% higher than the decadal scale estimates of Conard and Ivanova [1997] and French et al. [2000]. It is important to recognize that the burn severity parameters for boreal North America in our simulations are based on burn severity parameters from French et al. [2000]. Amiro et al. [2001] used the Canadian Forest Fire Behaviour Prediction System (FBP) System model [Forestry Canada, 1992] to estimate both the surface and crown fuel consumed during a fire, and used these estimates to calculate carbon emissions for Canada. Although interannual variability in our emissions between 1959 and 1995 are highly correlated with those of Amiro et al. from 1959–1995 (Figure 8; R2 = 0.92), they are higher by about 50%. The discrepancy between our estimates and Amiro et al. [2001] appears to be associated with the higher level of soil organic matter consumed associated with our use of the French et al. [2000] carbon consumption estimates [see also Kasischke and Bruhwiler, 2002].

Figure 8.

Comparison of TEM emission estimates for Canada with estimates from Amiro et al. [2001].

Table 3. Comparison of Emissions Estimates (Total Carbon Emitted, Tg C yr−1) From Previous Studies With Estimates Developed in This Study
RegionStudyYearsEmissionsThis Study
With CO2Without CO2
  • a

    Range is based on average emissions from low and high burn severity scenarios for this period.

CanadaAmiro et al. [2001]1959–1995264140
 Mouillot et al. [2006]1990–1999434947
Boreal North AmericaConard and Ivanova [1997]1971–1991 mean425553
 French et al. [2000]1980–1994 mean536160
 Conard et al. [2002]199852–558583
 van der Werf et al. [2006]1997–2002 mean354846
Pan-borealConard et al. [2002]1998187–245358349
 Kasischke and Bruhwiler [2002]1998290–383358349
 Kasischke et al. [2005]Range of mean emissions for 1996–2002a110–211262255
 Yurganov et al. [2004]1996–2001 mean6–63225219
 Zhuang et al. [2006]1990–1999 mean58256245
 Mouillot et al. [2006]1990–1999 mean209256245
Boreal Russia/SiberiaConard and Ivanova [1997]1971–1991 mean194204197
 Conard et al. [2002]1998135–190273266
 Shvidenko and Nilsson [2000]1988–1992 mean58244230
 Kajii et al. [2002]1998176273266
 Soja et al. [2004]1998–2002 mean116–520261252
 van der Werf et al. [2006]1997–2002 mean185223216
 Mouillot et al. [2006]1990–1999 mean166194185

[34] Across the pan-boreal region from 1996–2002, our estimates of emissions are higher than the range of emissions estimated by Yurganov et al. [2004], Kasischke et al. [2005], Mouillot et al. [2006], and Zhuang et al. [2006] (Table 3). Note that the range of emissions estimated by Kasischke et al. [2005] does not overlap with the range of Yurganov et al. [2004]. Our estimates of fire emissions for boreal Russia (Table 3) are also higher than those of Conard and Ivanova [1997] for 1971–1991 and those of Shvidenko and Nilsson [2000] for 1988–1992, which are time periods that correspond to the backcasting portion of our simulations for Eurasia. Our estimates are also higher than those estimated by van der Werf et al. [2006] for the period 1997–2000 and by Mouillot et al. [2006] for the 1990s. In contrast, our estimate of fire emissions for boreal Siberia from 1998–2002 are within the range reported by Soja et al. [2004, Table 3], but it should be noted that the range is quite large.

[35] Because 1998 was a high fire year in Eurasia, a number of studies have conducted analyses of fire emissions for that year. Our estimate of 1998 emissions at the pan-boreal scale (Table 3) is within the range reported by Kasischke and Bruhwiler [2002], but is substantially higher than the range reported by Conard et al. [2002]. Similarly, our estimates at the boreal Russia/Siberia scale (Table 3) are substantially higher than Conard et al. [2002] and Kajii et al. [2002].

[36] The comparison of fire emissions between our study and those of other studies identify that there is substantial uncertainty in estimates of fire emissions in the pan-boreal region. Our estimates of fire emissions tend to be higher than many of the previously published estimates because of the burn severity parameters that we used in this study. Thus, the uncertainty among studies appears to be largely associated with how burn severity is implemented among the approaches, an issue which we discuss in more detail below.

4.2. Comparison of Carbon Balance Estimates

[37] Inverse modeling studies have estimated exchange of CO2 between the pan-boreal region and the atmosphere based on variability in the concentration of CO2 that has been measured at various sites throughout the globe [e.g., Dargaville et al., 2006]. The results of our simulations for the combined responses to changes in atmospheric CO2, climate, and fire are within the range of uncertainty reported by Gurney et al. [2004] for boreal Asia and boreal North America from 1992–1996 (Table 4). However, it is important to note that the range of uncertainty from inversion-based modeling studies is quite large. We further compare our results to inventory- and process-based modeling studies to gain additional insight. In interpreting these comparisons it is also important to recognize that our simulations only considered one disturbance factor (fire), and that other disturbance factors in the pan-boreal region (e.g., insect disturbance, forest harvest, and land-use change) have the potential to influence regional carbon dynamics. For example, the analysis of Kurz and Apps [1999] indicates that insect disturbance was responsible for more loss of carbon than fire from Canadian forests in the late twentieth century.

Table 4. Comparison of Previous Carbon Balance Estimates (Tg C yr−1) With Estimates From This Study
Study TypeYearsRegionLiterature EstimatesThis Study
With CO2Without CO2
  • a

    Average net carbon storage in vegetation only.

  • b

    Average net carbon storage in vegetation and soil while also taking into account fluxes generated by disturbances.

  • c

    Results include disturbances due to fire, insects, and logging.

Atmospheric inversion
Gurney et al. [2004]1992–1996Boreal North America Boreal Asia−200 ± 280 360 ± 51091 22712–52
Inventory-based
Shvidenko and Nilsson [2002]a1961–1998 meanRussia210 ± 3015970
Shvidenko and Nilsson [2003]b1961–1998 meanRussia32222068
Myneni et al. [2001]1995–1999 meanCanada Eurasia73 47080 3140.4–4
Kurz and Apps [1999]c1970–1989 meanCanada525812
Process-based
Chen et al. [2000]c1980–1996Canada53 ± 2757−1

[38] Inventory-based modeling studies capture a wide range of impacts on carbon dynamics from human to natural disturbance. These studies generally focus on particular transects or regions in the boreal forest, and are useful because they incorporate natural and anthropogenic disturbance regimes. In contrast, the estimates from our simulation consider the influence of fire disturbance in addition to CO2 fertilization and climate variability. In comparison to previous inventory studies for Russia, the increase in vegetation carbon storage estimated by our simulations are substantially lower than the increases estimated by Shvidenko and Nilsson [2002, 2003], which considered a broader array of disturbances (Table 4). Myneni et al. [2001] conducted a study that relied on regression relationships between satellite-derived reflectance and forest inventory information to estimate changes in carbon storage for terrestrial areas north of 30°N from 1995–1999. In comparison to estimates of Myneni et al. [2001], the estimates of changes in carbon storage from our simulations that incorporated atmospheric CO2 fertilization are slightly higher for Canada and substantially lower for Eurasia. Kurz and Apps [1999] conducted an inventory-based modeling study across Canada that analyzed variability across multiple decades. Over the period 1970–1989, they report a change in carbon storage that is similar to the estimate from our simulations that incorporate atmospheric CO2 fertilization (Table 4). They also reported that Canadian forests acted as a sink from 1920–1979, and then changed to a source from 1980–1989 as a result of changes to the disturbance regime as the area affected by insect outbreaks and fires increased in the 1970s. Our results are consistent with these conclusions, but it is likely that our estimates of carbon storage in Canada in the late twentieth century would be lower if we considered insect disturbance in addition to fire. However, carbon storage in our simulations would likely be higher if we considered the effects of nitrogen deposition in fertilizing ecosystems in eastern Canada. In general, the simulations we conducted that considered an atmospheric CO2 fertilization effect appear to be more consistent with estimates of changes in carbon estimated by inventory-based modeling studies than the simulations that excluded CO2 fertilization. Thus, our study suggests that ecosystem responses to changes in atmospheric CO2 may be important to consider in addition to changes in climate and disturbance regimes.

[39] The influence of fire has been incorporated into several process-based models and studies have focused primarily on either modeling the regional or global area burned [Venevsky et al., 2002; Thonicke et al., 2001] or investigated carbon fluxes in response to fire for specific regions [Chen et al., 2000, 2003; Hicke et al., 2003; Amiro et al., 2000; Peng and Apps, 1999]. While several process-based models have been applied at large spatial scales [Potter et al., 2005, 2003b; McGuire et al., 2001], they do not coincide well with our study region or have not explicitly considered the effects of historical fire. Zhuang et al. [2006] simulated the effects of fire on carbon dynamics for high-latitude ecosystems north of 50°N from 1860–2100 and reported an overall net CO2 source of 240 Tg C yr−1 for the 1990s. The approach of Zhuang et al. [2006] differed from the approach of this study in several ways, but the key methodological difference responsible for the differences in results of the two studies is the assumption by Zhuang et al. [2006] of a fixed fire return interval (150 years) throughout the region to account for fires prior to the start of the historical record. This highlights the sensitivity of simulated carbon dynamics to factors affecting the stand age distribution of forests in the simulations. Another process-based modeling study that has considered historical fire is Chen et al. [2000], which used the Integrated Terrestrial Ecosystem Carbon-budget model (InTEC) to simulate the annual carbon balance of Canada's forests from 1896–1996 in response to CO2, climate, nitrogen deposition, and disturbance (insects, logging, and fire). The analysis of Chen et al. [2000], which considered Canada as being one spatial unit, estimated that Canada was a sink for carbon from 1980–1996. Our simulations driven by changes in CO2, climate, and fire are within the range of variability reported by Chen et al. [2000] but are substantially lower for the simulations that excluded the effect of atmospheric CO2 fertilization (Table 4). While our analysis is not exactly comparable to Chen et al. [2000] as it did not consider the effects of forest harvest, insect disturbance, or nitrogen deposition, both studies highlight the potential importance of responses of ecosystems to variability in atmospheric CO2 and climate in addition to changes in disturbance regimes.

4.3. Relative Roles of CO2, Climate, and Fire

[40] The advantage of using process-based models for simulating carbon dynamics is that individual processes that affect carbon storage can be isolated. This helps to provide a better picture of the roles of different environmental factors on carbon storage that cannot be addressed through atmospheric inversion and inventory-based modeling studies. Our analysis identifies that CO2, climate, and fire each have substantial influences on simulated carbon dynamics across the pan-boreal region. For the factors included in this analysis, if we group the effects into nondisturbance factors (CO2 fertilization and climate variability) and disturbance factors (fire), our analysis indicates that the nondisturbance factors are primarily responsible for the estimated carbon sink for the period 1996–2002. A similar conclusion was found across Canada for the 1980s and 1990s, which is also consistent with other findings for that region [Chen et al., 2003]. As stated earlier, it is important to recognize that our simulations do not incorporate other disturbance factors including insect disturbance, forest harvest, and land-use change.

[41] Although the response of TEM to increases in atmospheric CO2 is highly constrained by the representation of the nitrogen cycle in the model [McGuire et al., 1993, 1997, 2001; Kicklighter et al., 1999], the model does have the capacity for a response to increasing atmospheric CO2 as the ratio of vegetation carbon to nitrogen widens [McGuire et al., 1997]. For the pan-boreal region from 1996–2002, the CO2 fertilization effect in our simulations is 7.5 g C m−2 yr−1. There is still substantial debate about whether or not CO2 fertilization is occurring in the terrestrial biosphere [Caspersen et al., 2000; Hungate et al., 2003; Luo et al., 2004; DeLucia et al., 2005; Norby et al., 2005], and the resolution of this issue remains an important challenge as many process-based models indicate that this factor is responsible for substantial sink activity in the terrestrial biosphere in recent decades [e.g., McGuire et al., 2001]. The comparison of our simulations that both incorporate and exclude atmospheric CO2 fertilization highlight this uncertainty. In general, the results of our simulations that incorporate an atmospheric CO2 fertilization effect appear to be more consistent with previous analyses of carbon dynamics in the pan-boreal region.

[42] The positive response of carbon storage to warming climate in our simulations is largely associated with the increase of soil nitrogen availability to vegetation as increased decomposition in response to soil warming enhances nitrogen mineralization. This response of TEM is well-documented [e.g., McGuire et al., 1992; Melillo et al., 1993; Xiao et al., 1997; Tian et al., 1999], but there is much interannual and spatial variability in the response as it depends on soil moisture status [McGuire et al., 2000a; Thompson et al., 2006]. Over decadal timescales the response to a warming climate in the simulation results reported in this study was in general characterized by a faster increase in net primary production (NPP) than in decomposition, a pattern that resulted in a carbon sink of 3.6 g C m−2 yr−1 associated with climate variability between 1996 and 2002 at the pan-boreal scale. The increase in NPP in our simulations is consistent with a number of studies that have suggested that NPP in the pan-boreal region has been increasing in recent decades in response to warming [Nemani et al., 2003; Euskirchen et al. 2006; Kimball et al., 2006, 2007; see Goetz et al., 2005]. Our study is also consistent with a recent study indicating that boreal ecosystems sequester more carbon in warmer years [Chen et al., 2006].

[43] Although the effects of nondisturbance factors generally outweigh the effects of fire, we show that it is important to incorporate the role of fire when calculating the overall carbon budget for the pan-boreal region. Incorporating fire in our analysis shows that it reduces carbon storage across the pan-boreal region and, in large fire years (or averaged over decades), can switch from acting as a carbon sink to a carbon source to the atmosphere. Thus, fire plays an important role in the interannual variation in source/sink relationships. Although we find that the effects of fire are less than the effects of CO2 and climate, increases in fire frequency and burn severity in a changing climate may enhance the effect of fire on carbon dynamics across the pan-boreal region.

4.4. Limitations, Uncertainties, and Future Challenges

[44] We encountered several issues when attempting to evaluate the role of historical fire on high latitude carbon dynamics. We identify four main challenges that are important in influencing fire emissions estimates as well as the overall carbon budget: (1) the length of historical fire records, (2) the methods used for calculating stand age distribution prior to the start of the historical record, (3) accurately representing the influence of burn severity on carbon and nitrogen consumption, and (4) the role of peatland fires in estimating fire emissions.

[45] The lack of long-term spatially explicit fire data for Eurasia continues to be a problem for attempting to evaluate the role of fire in carbon dynamics of this region. This limitation also creates a great challenge with respect to accurately representing the state of the fire-driven landscape through the insertion of fires prior to the short historical record using coarsely interpolated fire return intervals. Our results presented here rely on a seven year historical period in terms of inserting pre-historical fires and therefore the frequency and size of fires in the short period is most likely not representative of the long term dynamic of fires that occur across Eurasia. Extending the existing satellite derived fire record prior to 1996 would help to reduce uncertainties. The extensive historical fire record for North America gives us a better understanding of the role of fire on carbon dynamics over the longer term and can be used to help reduce uncertainties that are associated with the short record for Eurasia for interpolating fire return intervals (e.g., standardizing Eurasian FRIs relative to North American FRIs).

[46] Another challenge that is closely related to the issue of data limitation on historical fires is the need to accurately represent the age distribution of forests prior to the start of using historical fire records in simulations. McGuire et al. [2004] documented that assumptions about historical fire prior to the start of the historical record have a large effect on simulations of carbon storage in Alaska. In boreal North America, we relied on using FRI based on the fire records from 1950–2002 in Alaska and 1959–2002 in Canada. The implementation of this approach essentially makes the assumption that the fire effect is neutral over these time periods in Alaska and Canada. However, our simulations estimated a fire effect of 15.6 to 17.4 Tg C, depending on CO2 fertilization (Table 2), from terrestrial ecosystems of boreal North America to the atmosphere. Thus, the fire effect we report for boreal North America in this study may largely be an artifact of how fires were inserted prior to the start of the historical record. For Eurasia we relied on using FRI from sparse literature estimates, which may result in estimates of FRI that are not entirely representative of a given region or of a particular vegetation type. The limitations imposed by available data for this region further compounds the problem in that the pre-historical fires are inserted based on a seven year burn record for Russia. The comparison between the results of Zhuang et al. [2006] and this study highlight the need for spatially explicit data sets on stand-age distribution in order to evaluate methodologies that estimate stand-age distributions prior to the start of historical fire data. It should be recognized that if stand age has been the result of multiple disturbances in a region, then the reconstruction of stand age distributions will need to consider the relevant set of disturbances in the region.

[47] A third challenge to incorporating fire into carbon balance estimates is related to the aboveground and ground layer carbon fraction consumed during a fire. Currently the definition of aboveground and ground layer carbon consumption and differentiating between fire regimes is limited to our understanding of what is presented in the literature and can therefore be taken only as a coarse estimate of what might actually be occurring in a given region. Also, the consumption parameters that we implemented in this study are fixed in time and do not take into account the seasonal variation in depth of burn. The importance of accounting for depth of burn is highlighted by Kasischke et al. [2005] and Kasischke and Turetsky [2006]; however, accounting for these seasonal differences in depth of burn will require that relationships among burn severity, seasonality of fire, and other factors be developed.

[48] An issue related to burn severity is the amount of nitrogen combusted from soil and vegetation nitrogen pools at the time of fire. Our assumption of 15% nitrogen loss from soil and vegetation at the time of fire is based on the assumption that nitrogen loss is highly variable across the boreal forest and in some cases can be reintroduced to the system by canopy ash [Harden et al., 2004]. We conducted a sensitivity analysis that evaluated this uncertainty by assuming no retention of soil and vegetation nitrogen at the time of fire [see Wang et al., 2001]. We found that across the pan-boreal region from 1996–2002, the effect of fire on carbon storage increased (i.e., became more of a source) by a factor of 50%, and decreased the overall carbon sink in response to all factors by 7%. Thus, in addition to better information on how burn severity influences carbon release, information on how burn severity influences nitrogen release would help improve the ability to represent interactions between how carbon and nitrogen affect carbon storage.

[49] The fourth challenge to incorporating fire into carbon balance estimates of the pan-boreal region is the role of peatland fires. Several studies have highlighted the importance of peatland fires in calculating current and future fire emissions [Turetsky et al., 2002; Kasischke and Turetsky, 2006; Turetsky et al., 2006]. With projections that some high-latitude regions will become drier in addition to warmer, it is possible that the fire regime will shift to later in the growing season, which may result in greater peatland fuel consumption with deeper thaw depths and therefore higher fire emissions. Therefore, it is important to accurately represent peatland burning in future studies to reduce uncertainties associated with estimating fire emissions.

5. Conclusion

[50] Our analysis suggests that CO2, climate, and fire each are important in the carbon dynamics of the pan-boreal region at interannual, decadal, and multidecadal timescales. It also shows that it is important to incorporate fire in a temporally and spatially explicit manner when simulating the effects of fire on carbon dynamics for the boreal forest. While our analysis does not consider the full suite of disturbances that occur in the pan-boreal region, our estimates are generally within the uncertainty of those presented in previous inversion-, inventory-, and process-based modeling studies within this region.

[51] Our analysis indicates that fire plays an important role in the interannual and decadal scale variability of source/sink relationships of the pan-boreal region. Other analyses indicate that changes in fire regime have the potential to substantially influence carbon source/sink relationships of northern terrestrial ecosystems at multidecadal to century timescales [McGuire et al., 2004; Zhuang et al., 2006]. While we found that the pan-boreal region acted as a carbon sink for the period 1996–2002 in response to CO2, climate, and fire, with Eurasia accounting for more than half of this reported sink activity, fire in this time period tended to decrease the strength of the sink. Although we report that the pan-boreal region is currently acting as a net carbon sink when considering changes in atmospheric CO2, climate and fire, there are substantial uncertainties in our estimates. These uncertainties are due to several factors which include sparse fire data across the Eurasian continent, uncertainty in estimating carbon consumption, and the difficulty in verifying assumptions about the representation of fires that occurred prior to the start of the historical fire record. The reduction of these uncertainties can be accomplished through the retrospective extension of the satellite-derived burn record in Eurasia back to the early 1980s using existing methods, better information on the spatial and temporal variability of above- and below-ground carbon fraction consumed, and the spatially explicit representation of stand age distribution throughout the pan-boreal region.

[52] It is currently difficult to project what the combined effects of increasing atmospheric CO2, a changing climate dominated by increasing temperatures, and a changing fire regime dominated by increased burn severity, frequency, and size of fire will have on net carbon storage across boreal North America and Eurasia. If the proportion of large, severe fires increase as a result of a warmer climate, then the sink strength of northern terrestrial ecosystems may be weakened and potentially switch to becoming a carbon source to the atmosphere. Our ability to project future temporal and spatial changes in carbon dynamics at large spatial scales is limited by our understanding of how the temporal and spatial aspects of fire influence historical carbon dynamics. Further analyses of the retrospective role of fire in the pan-boreal region should include (1) improved data sets of fire area for Eurasia, (2) improved estimates of how carbon consumed by fire varies spatially and temporally, and (3) integration of fire with other important disturbances so that reconstructions of stand age based on assumptions about historical disturbance can be verified with data on current stand-age distributions.

Acknowledgments

[53] Funding for this study was provided by grants from the National Science Foundation Biocomplexity Program (ATM-0120468) and Office of Polar Programs (OPP-0531047 and OPP-0327664); the National Aeronautics and Space Administration Land Cover Land Use Change Program (NAF-11142) and North America Carbon Program (NNG05GD25G); the Bonanza Creek LTER (Long-Term Ecological Research) Program (funded jointly by NSF grant DEB-0423442 and USDA Forest Service, Pacific Northwest Research Station grant PNW01-JV11261952-231); and the U.S. Geological Survey. This study was also supported in part by a grant of HPC resources from the Arctic Region Supercomputing Center at the University of Alaska Fairbanks as part of the Department of Defense High Performance Computing Modernization Program.

Ancillary