Correspondence: Scott D. Peckham, tel. 608 265 5628, fax 608 262 9922, e-mail: firstname.lastname@example.org
Forests of the Midwestern United States are an important source of fiber for the wood and paper products industries. Scientists, land managers, and policy makers are interested in using woody biomass and/or harvest residue for biofuel feedstocks. However, the effects of increased biomass removal for biofuel production on forest production and forest system carbon balance remain uncertain. We modeled the carbon (C) cycle of the forest system by dividing it into two distinct components: (1) biological (net ecosystem production, net primary production, autotrophic and heterotrophic respiration, vegetation, and soil C content) and (2) industrial (harvest operations and transportation, production, use, and disposal of major wood products including biofuel and associated C emissions). We modeled available woody biomass feedstock and whole-system carbon balance of 220 000 km2 of temperate forests in the Upper Midwest, USA by coupling an ecosystem process model to a collection of greenhouse gas life-cycle inventory models and simulating seven forest harvest scenarios in the biological ecosystem and three biofuel production scenarios in the industrial system for 50 years. The forest system was a carbon sink (118 g C m−2 yr−1) under current management practices and forest product production rates. However, the system became a C source when harvest area was doubled and biofuel production replaced traditional forest products. Total carbon stores in the vegetation and soil increased by 5–10% under low-intensity management scenarios and current management, but decreased up to 3% under high-intensity harvest regimes. Increasing harvest residue removal during harvest had more modest effects on forest system C balance and total biomass removal than increasing the rate of clear-cut harvests or area harvested. Net forest system C balance was significantly, and negatively correlated (R2 = 0.67) with biomass harvested, illustrating the trade-offs between increased C uptake by forests and utilization of woody biomass for biofuel feedstock.
Forests play a significant role in the global carbon cycle (Schlesinger, 1997; Pan et al., 2011), accounting for 67% of terrestrial carbon dioxide uptake from the atmosphere (Landsberg & Gower, 1997). Forests in North America are an important source for wood fiber for the wood and paper product industries (Williams, 1989; Birdsey et al., 2006; Gower et al., 2006) and these same forests sequester carbon (C) from the atmosphere (Goodale et al., 2002; Crevoisier et al., 2010). These forests also provide valuable habitat for resident and migratory animals, especially neotropical songbirds. Forests in the Midwest region of the United States contain over 3 billion m3 of wood (Smith et al., 2010) and therefore are a potential source for biofuel feedstock (Mabee et al., 2011). Harvest residue, which can comprise up to 35% of aboveground biomass (see Peckham & Gower, 2011), is the most likely source of feedstock (Mabee et al., 2011), but other sources include nonmerchantable trees (e.g., small diameter material, cull trees, etc.) or an increase in harvest area or intensity. Although the supply of woody biomass is not likely sufficient to replace the majority of fossil fuel energy use in the United States or globally (Converse, 2007) and the benefits of increased biomass harvest may vary by region (Van Deusen, 2010; Hudiburg et al., 2011), it could reduce the consumption of fossil fuel and associated carbon dioxide (CO2) emissions to the atmosphere (Schlamadinger & Marland, 1996). Few studies have attempted to quantify the woody biomass resources in the Midwest United States that could be available for different harvest scenarios and the trade-offs between biofuel production, carbon storage in forests, and net fluxes of carbon.
Before any large woody biomass harvest programs are initiated it is important to study the entire forest system carbon balance. For example, is the benefit of an increase in woody biomass feedstock offset by a similar or greater decrease in carbon sequestration by the biological ecosystem? It remains unclear how the anticipated increase in utilization of the woody biomass for biofuel feedstock and bioenergy production will affect the whole-system carbon balance. The forest system is composed of the biological and industrial carbon cycles (Gower, 2003). Although there is great interest in using woody biomass as a biofuel feedstock, there is woefully inadequate information on the whole-system carbon balance, especially the feedbacks of increased biomass removal on long-term soil fertility (Peckham & Gower, 2011). One potential impact that has received much attention is the change in total carbon dioxide emissions related to the production and use of biofuels. Because the biological and industrial C cycles are linked through harvesting and management practices, a whole-system approach, as originally proposed by Odum (1969), is needed to model the forest system carbon cycle (Gower, 2003). Landscape-level effects of management and product choices on the future forest (biological and industrial) C balance remain largely unknown.
Ecosystem process models such as Biome-BGC (Running & Gower, 1991; Thornton et al., 2002; Bond-Lamberty et al., 2007b) simulate the effects of disturbance, such as forest harvesting, on the biological C cycle (Petritsch et al., 2007; Peckham & Gower, 2011; Peckham et al., 2012, 2013), and have been applied over large geographical regions (Bond-Lamberty et al., 2007b; Peckham et al., 2012, 2013). Unlike empirical growth and yield models, process-based ecosystem models simulate water, nitrogen, and their interaction with the carbon cycle, as well as potential adverse effects of increased biomass removal on nitrogen availability and its effects on forest productivity (Peckham & Gower, 2011). In the industrial ecosystem, forest product life-cycle inventories (LCI, a discrete step in the life-cycle assessment process) are used to determine the environmental burden of a product or process (Gower, 2003; White et al., 2005; Gower et al., 2006). LCI's have been developed to estimate greenhouse gas or carbon dioxide equivalent emissions (CO2-eq) for the forest harvesting process (White et al., 2005), production of dimensional lumber (White et al., 2005; Gower et al., 2006; Bergman & Bowe, 2010), oriented strand board (OSB) (White et al., 2005), and magazine paper (Gower et al., 2006) in the Midwest region. Coupling the biological and industrial models simulates the whole-system C cycle of management practices and product production and their impacts on forest sustainability, future forest growth at regional-to-landscape scales, and can elucidate forest management effects on forest system carbon dynamics.
Whereas most life-cycle studies of biofuel and bioenergy assume C neutrality (see Cherubini et al. (2011)), there is evidence that this approach is inadequate (Rabl et al., 2007; Johnson, 2009). Moreover, these studies do not account for feedbacks associated with increased harvest and residue removal on the C dynamics of the biological ecosystem (Peckham & Gower, 2011). In this study we simulate the biological and industrial ecosystems beginning with C uptake from the atmosphere by trees through emissions (both biogenic and fossil fuel) associated with the production, use, and disposal of common products and biofuel. This approach includes feedback of forest management on the biological system and the influence of product production on the overall C balance of the forest (biological + industrial) system.
The overall objective of this study was to simulate the available woody biomass feedstock and its effects on whole-system carbon balance of deciduous and conifer forests in the Midwest for different forest harvest and biofuel production scenarios. Yearly estimates of net biome production (NBP, net ecosystem production integrated over space (Chapin et al., 2006), or annual rate of C sequestration) and harvested biomass were used as input to the industrial C cycle model. We hypothesized that increasing both forest harvest and biofuel production would increase the industrial C budget and decrease NSP in both hardwood and conifer forests of the Midwest region.
Materials and methods
The modeling region was the forested area within the boundary of the Midcontinent intensive (MCI) study area of the North American Carbon Program (http://www.nacarbon.org/nacp/mci.html) (Fig. 1). This area includes the states of Minnesota, Wisconsin, Iowa, Illinois, and portions of North Dakota, South Dakota, Nebraska, Kansas, Missouri, Indiana, and Michigan. The entire MCI region encompasses 125 × 106 ha of forest, agriculture, and urban landscapes. Temperate forests comprise roughly 18% of the total area and consist primarily (75%) of deciduous broadleaf (hardwood) forests. The second most common forest type is evergreen needleleaf (conifer) forests in both uplands (Pinus spp.) and lowlands (Picea spp.). Forests are the dominant vegetation cover in northern Michigan, Wisconsin, and Minnesota, and in northern Missouri and southern Illinois. The terrain is generally low relief (Potter et al., 2007), with some rolling hills and deep river valleys. Climate ranges from long, cold winters and a short growing season (<120 days) in the northern regions to mild winters and long (>180 days) growing season in the southern region.
Biological system model
The biological forest C cycle was modeled using Biome-BGC, an ecosystem process model with interacting carbon (C), nitrogen (N), and water cycles. Biome-BGC runs on a daily time step, and requires daily minimum and maximum air temperature, total solar irradiance, average vapor pressure deficit (VPD), and total precipitation. Photosynthesis, evapotranspiration, and respiration are calculated for both sunlit and shaded portions of the plant canopy. Photosynthesis is computed using the Farquhar model (Farquhar et al., 1980) and evapotranspiration is calculated by the Penman–Montieth equation (Monteith, 1965; Campbell & Norman, 1998). The model includes decomposition fluxes from litter and soil organic matter (SOM). Litter is characterized by its chemical composition, which is part of the input parameter set. SOM is modeled as a cascading system of four pools of increasing recalcitrance (Thornton et al., 2002). The decomposition flux (i.e., heterotrophic respiration) of the litter and SOM pools is dependent on the rate constant, the soil temperature and moisture, and the size of each pool. Woody debris is subject to physical degradation before entering the litter pool. Both plant uptake and decomposition processes compete for the same pool of mineral nitrogen, which occur at reduced rates when nitrogen demand exceeds supply. Complete model logic and physiological processes have been described in detail previously (Running & Coughlan, 1988; Running & Gower, 1991; Kimball et al., 1997; White et al., 2000; Thornton et al., 2002), including sensititvity to both environmental drivers and ecophysiological parameters. The model version used in this study is capable of simulating large geographical regions (Bond-Lamberty et al., 2007b, 2009; Peckham et al., 2012, 2013), disturbance (Bond-Lamberty et al., 2007b, 2009; Peckham & Gower, 2011; Peckham et al., 2012, 2013), and includes improvements for flooded soils including oxygen-limited decomposition (Bond-Lamberty et al., 2007a). Model simulations within the MCI region have been previously compared with field measurements of soil C, NPP, and NBP for even-aged and uneven-aged forests in northern WI (Peckham & Gower, 2011; Peckham et al., 2012), and well and poorly drained boreal stands comprising a disturbance chronosequence (Bond-Lamberty et al., 2005). Biome-BGC simulation of soil C accumulation in a sugar maple stand compared favorably with data measured by Tang et al. (2009) and has been discussed previously (Peckham & Gower, 2011). We have outlined in great detail Biome-BGC's performance simulating the major C stores and fluxes under both natural and human disturbances, including multiple-model comparisons, in a previous study (Peckham & Gower, 2011). We observed good agreement between measured and simulated NPP for two forest types in northern WI (Peckham & Gower, 2011), measured and modeled NPP for well- and poorly drained wildfire chronosequences (Bond-Lamberty et al., 2007a), and measured and modeled soil C accumulation, NPP and NEP for northern hardwood chronosequence (Peckham & Gower, 2011; Peckham et al., 2012, 2013).
The near present-day (2004) conditions in Biome-BGC were estimated using a two-step initialization procedure. First, a spin-up or model self-initialization (Thornton & Rosenbloom, 2005) was run using 25 years of historical meteorological data (1948–1973) and preindustrial estimates of atmospheric carbon dioxide (CO2) concentration and nitrogen deposition (Ndep), 280 ppm and 0.001 kg m−2, respectively. Biome-BGC does not simulate land cover conversion (i.e., forest to agriculture or vice versa) and hence we only simulated the current forested area and assumed those areas now forested have been so since the beginning of the simulation. We assumed that prior to 1800 the forest was intact, primary, only subject to natural disturbances, and that NBP was in relative equilibrium with the atmosphere (i.e., C balance approximately zero). These assumptions are consistent with historical accounts in the region (Williams, 1989; Whitney, 1994). Wind and fire disturbances are rare (>1000-year return interval) in hardwood forests, but more frequent (50–200-year return interval) in conifers (Frelich, 1995). The mortality fractions in Biome-BGC were set to reflect these disturbances using data available from the region (Whitney, 1994; Frelich, 1995; Cleland et al., 2004).
Following spin-up, we simulated the 1800–2004 period to estimate near present-day model C and N pool values and initialize all of the harvest simulations in this study. Much of the MCI's forest landscape was clear cut in the 19th century; therefore, a historic disturbance data set was constructed to simulate past land-use history between 1800 and 2004. Starting with estimates of the current stand age (approximately 2004) for each cell and working backwards, we created a general disturbance regime and simulated it over 1800–2004 to estimate current model C pools for each grid cell. For the model initialization simulation, we assumed all current stands originated from a harvest, and each location was harvested at least once, with the exception of areas within the Boundary Waters Canoe Area (MN), Porcupine Mountains (MI), and Sylvania Lakes (MI) Wilderness areas, which are the only remaining old-growth forests of significant area in this region (Frelich, 1995). We simulated two types of harvest, clear cut and selective (or partial cut). Clear-cut harvest assumed 100% removal of trees, and vegetation growth was subsequently restarted. Selective harvests were assumed to remove 25% of the canopy area and biomass. The proportion of selective harvest (of the total area harvested annually) was assumed to increase linearly from 0.0 in 1930 to 0.75 in 2004, reflecting the increasing use of selective harvest from its adoption in the early- to mid-1900s (Gronewold et al., 2010) to present day (Smith et al., 2010). The resulting age structure in 2004 matched FIA data. Atmospheric CO2 concentration and atmospheric nitrogen deposition (Ndep) varied from preindustrial (1800) estimates up to near present-day (1998) levels. Complete description of the model initialization and management simulations are given in Peckham et al. (2013), and a similar initialization routine was also described in Peckham et al. (2012).
The model simulated the current forested areas in the MCI region (approximately 205 000 cells) at a spatial resolution of 1 km2. The forest type of each pixel was determined using a data product derived from US Forest Service forest inventory and analysis (FIA) data (Wilson et al., 2009-ongoing), an extension to a gradient nearest neighbor method (Ohmann & Gregory, 2002). This method integrates FIA plot, remote sensing, and other environmental data to predict forest cover (and other variables) on a 250 m grid. This gridded data set was provided by the US Forest Service Northern Research Station and then resampled to match our 1 km2 modeling grid. We assumed that forest area did not change during the 50-year simulation. Other spatial data, including soils, elevation, climate, and vegetation ecophysiological parameters used to drive Biome-BGC simulations, and the full model initialization are detailed in Peckham et al. (2013).
Daily climate data from the NCEP reanalysis project (http://www.esrl.noaa.gov/psd/data/gridded/reanalysis/) were used to drive the model. The climate data from 1948 to 2004 were resampled to a 50 km2 resolution grid (climate data only) that matched the spatial extent of the common modeling region following techniques used in previous modeling studies (Bond-Lamberty et al., 2007b, 2009; Peckham et al., 2012). This resolution captures the broad climate patterns in the region. The focus of this study was to elucidate the effects of harvest scenarios, not to simulate climate change effects on forest growth. An additional challenge is that the available climate data do not cover the entire time period of interest. Therefore, we elected to focus solely on harvest effects and remove the effects of interannual climate effects on the disturbance (harvest) response in each grid cell by using a climate ensembling method (Thornton et al., 2002). The biological system model outputs (and industrial system input) were an average of 50 individual simulations, each simulation beginning with a different year in the climate data record. Due to the extreme amount of data generated by the ensembling method, only a mean value of the 50 ensemble simulations for each grid cell was produced in each model run (i.e., variability due to climate was not retained). Most recent wet and dry deposition rates of nitrogen were obtained from the National Atmospheric Deposition Program (http://nadp.sws.uiuc.edu/) and held constant throughout the simulation.
Industrial carbon cycle model
The biomass (carbon) from simulated harvests in the biological model (Biome-BGC) was the input to the industrial ecosystem model (Fig. 2). The model computed the total carbon dioxide equivalent greenhouse gas emissions (CO2-eq) associated with the harvest, transportation, use, and disposal of the dominant wood products in the modeling region following Gower et al. (2006). The model specifically accounts for the production, use, and final fate of dimensional lumber, OSB, magazine paper, and biofuel. Because of its prevalence in regional harvest, the domestic burning of wood for heat (fuelwood) is also included for both hardwood and conifer forests. Except for the biofuel and wood burning LCIs, all the data were obtained from the MCI region. In the calculation of NSP and results presented later, we convert CO2-eq emissions to elemental C and report NSP and Et in g C m−2. The carbon losses from the production, use, and disposal of each product were computed using emission factors (g C emitted per g C processed) for harvesting operations, paper, and OSB [0.10, 0.16, and 0.12, respectively (Peckham et al., 2012)]. Dimensional lumber emission factors were 0.11 and 0.13 for hardwood and softwood, respectively (Peckham et al., 2012). The emission factor for domestic fuelwood (1.13) was computed as the mean value of the emission factors for fireplaces, EPA-approved wood stoves, and pellet stoves (Lee et al., 2010). For the production of biofuel an emission factor of 0.22 was used, based on the mean of four estimates for the production of ethanol from woody biomass in Norway (Bright & Stromman, 2009). We used this study because data for a facility in the United States were not available. Key assumptions used in the biofuel production LCI were a 4 : 1 use ratio of traditional roundwood to residues, baling of residues at the harvest site, and chipping at the refinery site. We used only the statistics for the conversion process at the biorefinery. Regional harvest and transportation emission factors listed above were applied prior to the harvested C reaching the biorefinery. Carbon dioxide emissions from combustion of ethanol in an E85 system (0.45 kg C l−1) in light-duty vehicles in the MCI region were computed using data generated by the GREET model presented in Timpe & Aulich (2005).
Forest management scenarios
The forest management scenarios simulated in this study were designed to examine the effects of harvest intensity (clear cut vs. selection), harvest residue removal rate (as a fraction of total biomass harvest), and total annual harvest area on the biological and industrial C fluxes in the MCI region. Because the study region spans multiple forest ownership types (i.e., federal, state, county, private nonindustrial, etc.), current levels of forest harvest were determined for selected regions within the MCI using disturbance data derived from long time series of Landsat data (Stueve et al., 2011) and for the entire MCI area using regional statistics (Smith et al., 2010). The management scenarios simulated in this study varied the total area harvested, the amount of clear-cut vs. selective harvest, and residue left on the site after harvesting. A brief description of the seven management scenarios is provided in Table 1, and complete description is provided by Peckham et al. (2013). On the basis of regional statistics, we consider our ‘base’ (B) simulation to represent current harvest levels (2004). Averaged for the entire study region, annual area harvest averaged 1.7% of the forested area (2001–2005) and clear-cut and partial (selective) harvest percentages were 25 and 75, respectively (Smith et al., 2010). The low-residue (LR) and high-residue (HR) scenarios were identical to the base, except that 15% and 35% of harvested biomass, respectively, was left on site compared to 25%. The next two scenarios either increased (IA) or decreased (DA) the area harvested by a factor of two relative to 2004 levels. The final two scenarios increased (IC) or decreased (DC) the percentage of clear-cut harvest type by 25% (and also changed the selective harvest, so the sum remained 100%).
Table 1. Details of the seven forest management scenarios simulated using Biome-BGC. See Peckham et al. (2013) for more detailed description of the scenarios and Peckham & Gower (2011) for details on harvest and residue retention simulation in Biome-BGC
Harvest (as a fraction current rate)
Clear cut (% of total)
Selective cut (% of total)
Residue retention (% of harvest biomass)
Base + low residue (LR)
Base + high residue (HR)
Decreased clear cut (DC)
Increased clear cut (IC)
Increased area (IA)
Decreased area (DA)
Biofuel production scenarios
In addition to the management scenarios simulated in the biological system model, the harvest biomass entering the industrial system model was simulated for three biofuel production scenarios: (1) no biofuel production, the current wood and paper product production statistics were used for all forest management scenarios, (2) production of biofuel and traditional wood products, where all excess harvest (traditional and residue) was used to produce biofuel for the three management scenarios with harvest amount that exceeded the base (only the IA, LR, IC management scenarios met this criteria), and (3) all harvested biomass from each of the seven management scenarios was allocated to biofuel feedstock. In this study, current wood and paper production statistics were obtained from the timber products output database (http://srsfia2.fs.fed.us/php/tpo2/tpo.php). The base production proportions (of total harvest) for lumber, OSB, paper, and fuelwood used were 0.37, 0.15, 0.30, and 0.18 for hardwood and 0.38, 0.47, 0.11, and 0.03 for conifers, respectively. These were used directly in the model when no biofuel was produced, but were modified to meet the criteria in the two scenarios where biofuel was produced.
Model outputs and interpretation of results
Net biome production (NBP), industrial emissions (ET), total forest carbon (sum of vegetation, litter, and soil), and net system production (NSP, NBP minus emissions) were calculated annually for each biofuel production scenario for the seven management scenarios (total of 17 model runs of 50 years each). Positive values of NSP indicate net C sequestration, negative values indicate net emission to the atmosphere. Harvest scenario simulations (Table 1) were compared with the base (B) simulation. Yearly rates of change, or departure from the base, were reported in g C m−2 yr−1 or as a percentage change over the 50-year simulation. Industrial emissions were also stratified by major forest type: deciduous broadleaf (hardwood) and evergreen needleleaf (conifer) forests. No formal statistics were performed on the model results because the sample size is so large (n = 206 000 cells) that all treatment effects were significant (i.e., standard errors were less than 0.5 g C m−2 yr−1). Therefore, we elected to express results relative to the base scenario.
Total forest carbon
Total ecosystem (biological) carbon for the MCI region at simulation start was 57 × 103 g C m−2 yr−1 (Fig. 3). The low-intensity management scenarios (DC, DA) allowed total C stores to increase by 9.1% and 10.5%, respectively, over the simulation period (Fig. 3). High-intensity harvest regimes IA and IC decreased C stores by 3.0% and 2.3%, respectively. The current harvest regime increased C stores by 5%.
NBP and harvest
Biome-BGC outputs of NBP for the MCI region ranged from 129 to 170 g C m−2 yr−1 (Fig. 4a) for the seven management scenarios. Harvest ranged from 44 to 161 g C m−2 yr−1 (Fig. 4b) and averaged 98 g C m−2 yr−1 for the seven scenarios. The more intensive harvests (IA and IC) had the lowest NBP whereas the least intensive harvests (DA and DC) had the largest NBP (Fig. 4c). Increasing residue removal (i.e., LR vs. B) increased average harvest by 13% and NBP by 9% (Fig. 4c), whereas increasing residue (HR) decreased harvest by 14% and NBP by 9%. NBP was negatively correlated with annual harvest of live biomass (−0.33 g C m−2 yr−1, P < 0.05). Although within a given level of live biomass removal, increasing removal of residue increased NBP.
The industrial C budget (Et) for the MCI region ranged from 17 to 97 g C m−2 yr−1 (Table 2a) averaged over the 50-year simulations. The IA and IC scenarios had the largest Et, whereas the DC and DA scenarios had the lowest in all biofuel and management scenarios. Producing only ethanol from forest harvest increased total emissions by 55–60% compared with current wood and paper production (no biofuel) for all management scenarios. Increasing harvest to supply biofuel production increased Et 23%, 103%, and 120% for LR, IC, and IA management options, respectively, compared with the base scenario. Producing biofuel instead with traditional products, increased total emissions by only 8 (LR) to 24% (IA) when compared with the same forest management scenario with no biofuel production (i.e., all harvest in IA, IC, LR was used to produce paper, lumber, OSB, and wood fuel).
Table 2. Industrial carbon (C) emissions from the seven simulated management strategies under three different biofuel production scenarios for all forests (a), hardwood (b), and conifer forests (c). Results shown are averages over the 50-year simulation period
Industrial emissions (g C m−2 yr−1)
Increased clear cut
Decreased clear cut
(a) All forests
Forest products + biofuel
(b) Hardwood forest
Forest products + biofuel
(c) Conifer forest
Forest products + biofuel
The industrial budget for hardwood forests (Eh) ranged from 19 to 107 g C m−2 yr−1 (Table 2b). Producing only biofuel increased Et by 49–53% in all forest management strategies. The IA and IC management scenarios had the largest industrial emissions. Increasing harvest to supply traditional wood products and biofuel production increased Eh 23%, 114%, and 134% for LR, IC, and IA management options, respectively, compared with the base scenario, and increased emissions by 7% (LR) to 23% (IA) when compared with the respective management scenario with no biofuel production.
Conifer forests in the MCI region generally had lower industrial C budget (Ec) than hardwoods (Table 2c). For the various management strategies and biofuel production scenarios, average Ec ranged from 9 to 60 g C m−2 yr−1. Producing only ethanol from forest harvest in conifer forests increased C emissions by 120–144% in all management scenarios. Increasing harvest area and intensity for biofuel production increased Ec by 29%, 100%, and 124% for the LR, IC, and IA scenarios, respectively, when compared with the base management strategy.
Net system production
For the MCI region, the 50-year mean NSP ranged from 33 to 152 g C m−2 yr−1 for the seven management and three biofuel scenarios. Over the simulation, NSP generally declined in each biofuel scenario (Fig. 5a–c). Annual NSP ranged from a minimum of −18 g C m−2 yr−1 to a maximum of 207 g C m−2 yr−1. The minimum NSP occurred under the increased area management when only biofuel was produced. Maximum NSP occurred when harvest area decreased and no biofuel was produced. Changing the biofuel production scenario from none to only biofuel decreased mean NSP from a minimum of 7% for the DA scenario to a maximum of 51% for the IA scenario. Producing wood products and biofuel decreased mean NSP by 2%, 18%, and 22% for the LR, IC, and IA forest management scenarios, respectively. In most scenarios, mean annual NSP was positive (i.e., net C sequestration) over the simulation period (Fig. 5a–c). However, for the IA scenario, NSP dropped below 0 g C m−2 yr−1 (net C source) when only biofuel was produced, approached 0 (i.e., C neutral) in the other two biofuel production scenarios, but increased toward the simulation end in all biofuel scenarios.
This study demonstrates the potential impacts that harvest and biofuel production have on the carbon cycle in existing forests of the Midwest United States, and the need to develop forest C management plans on whole-system analysis. Although less than 2% of the forest area is currently harvested annually region-wide, variations in harvest area and type significantly impacted the carbon balance of the forest system, in both biological and industrial cycles. The Upper Midwest region is characterized by a large and nearly contiguous forest in the north, a modest forest in the south, and scattered, local forests in the predominantly agriculture region in the center. To date, agriculture crops and residue have been the primary focus for biofuel feedstock. Depending on the harvest regime, we estimated that the MCI region could annually produce from 44 to 161 g C m−2 (roughly 18 to 66 × 106 t dry woody biomass). This value compares favorably with the 32 × 106 t dry residue biomass available in the southern United States (Eisenbies et al., 2009). However, scientists, land managers, and policy makers do not consider the effects of biomass removal on the carbon balance of the forests, and the whole system.
Effects of forest management on NSP and Et
Forest and residue harvest scenarios had a profound effect on simulated NSP. Annual model outputs ranged from a small C source (−18 g C m−2 yr−1) to a relatively strong C sink (207 g C m−2 yr−1) depending on management scenario (i.e., IA vs. DA) and biofuel production option (i.e., only biofuel vs. no biofuel production). The management scenarios produced harvests that differed by ±50% of the base, and hence had a large impact on total emissions, regardless of the biofuel production scenario chosen, simply due to more wood fiber entering the industrial C cycle. These simulations suggest that although only a small fraction of forested area is harvested annually (0.5–3.5%), the effect of harvest on the C balance of a stand, aggregated over a landscape, has the potential to make the regional NSP near or below zero in 25 years (e.g., IA scenario). The observed general decline in NSP for all management scenarios is tightly coupled to NBP (see Peckham et al., 2013 for detailed analysis of NBP) in this study for both hardwood and conifers (Fig. 5a–c), and is likely due to the increase in forest age, and its well-documented age-related NPP decline (Gower et al., 1996; Ryan et al., 1997). The general decline in NBP is also consistent with the projected trend for US forests in the next 100 years (Birdsey et al., 2006). In most scenarios simulated here, 75% or more of the harvests removed only 25% of the biomass, therefore a larger percentage of forest reach maturity although harvest frequency is relatively high in the region. This general trend is consistent with Birdsey et al. (2006), who reported a decrease in future C sequestration over US forests based on inventory data. They suggest that management practices could be implemented as a way to increase net C uptake by the US forest sector. The simulations in this study for the MCI region support their findings. Our estimate using current management and product production suggests that the forest system, as a whole, is still a C sink, and will continue to be, even when the production, use, and final fate of common forest products are included. However, our results also clearly show that increased harvest of biomass decreases the overall forest system C balance. These results highlight the need for a whole-system approach when developing a forest C management plan.
Implications of increased harvest residue removal and biofuel production
Increased harvest residue removal increased total (Et), hardwood (Eh), and conifer (Ec) emissions. However, in the 50-year period, NSP was greater in LR than HR scenarios. Changing product production to include biofuel decreased NSP in the LR management scenario (i.e., the difference between curve for LR in 5a and 5c). The increase in the industrial budget was not large enough to offset the relative increase in NBP in the LR scenario over the base (see Peckham et al., 2013 for NBP analyses). Although simulations here suggest that modest increase in utilization of harvest residue (+10%) does not negatively impact the C budget of the MCI region, previous model simulations at this level of residue removal (35%) repeated over several rotations decreased long-term soil fertility (Peckham & Gower, 2011). The simulation of harvest in Biome-BGC does not include any of the potential negative effects on soil physical properties (i.e., soil compaction, soil erosion, etc.) on forest regeneration and growth, which can be significant (Gent et al., 1984; Pye & Vitousek, 1985): therefore, these simulations should be viewed as conservative estimates.
Comparison with other studies
Simulations in this study suggest that biofuel production from conifer forests in the MCI region may decrease NSP (data not shown). Kilpelainen et al. (2011) found that integrating bioenergy along with timber production decreased net C exchange for two managed Norway spruce forest sites in Finland. Although their study utilized a different ecosystem process model and life-cycle inventories, their reported industrial C budgets (25–76 g C m−2 yr−1) generally agree well with our estimates.
Within the MCI region, our estimates of Et (17–40 g C m−2 yr−1) under current management (B) are larger than the 6–12 g C m−2 yr−1 reported by White et al. (2005) and the 4–12 g C m−2 yr−1 reported by Peckham et al. (2012) for the Chequamegon–Nicolet National Forest in northcentral Wisconsin. The discrepancy was indicated by the exclusion of burning of wood for fuel, which in this study was a major component of Et in hardwoods. The management regime in that region (Peckham et al., 2012) is more comparable with the decreased area scenario simulated here. Our estimate under the DA management scenario with no biofuel production for conifer forests of the MCI region (9 g C m−2 yr−1) is similar to these two previous studies.
The magnitude of the simulated total forest carbon (Fig. 3) is higher than typical values reported in the US Forest Service Inventory and Analysis (data not shown) and in estimates from previous decades (Houghton & Hackler, 2000). Greater than 50% of the simulated C store is in the soil layer, and simulated total vegetation C was less than 20 × 103 g C m−2 in 2004. In contrast, Houghton & Hackler (2000) report a typical value of roughly 9 × 103 g C m−2 in 1990 for broadleaf deciduous forests in the study region. Our estimate is not beyond vegetation biomass these forests once supported (Houghton & Hackler, 2000). Because both vegetation and soil carbon have a relatively long residence times in Biome-BGC, simulated biomass tends to be higher than other models (Wang et al. 2011). The values reported here are comparable with previous studies using this model within a similar climate space (Cienciala & Tatarinov, 2006; Wang et al. 2011). Despite the relatively high simulated biomass, our estimates of the industrial C budget are completely dependent on the harvest of simulated biomass. Harvesting a prescribed area of forest and processing the resulting biomass C yielded industrial C budgets that were comparable with other studies (White et al., 2005; Kilpelainen et al., 2011; Peckham et al., 2012). If simulated biomass in this study was excessively overestimated, these emission values would not be comparable.
Sensitivity of forest C budget estimates to all product emission factors
This study demonstrates the potential impact a single product or use can have on the industrial C budget. The current use of hardwoods for domestic fuel, about 20% of the hardwood harvest (Smith et al., 2010), significantly influenced total emissions. The emission factor (i.e., efficiency) used for burning wood for fuel (1.12) is up to an order of magnitude greater than the other products, which ranged from 0.09 to 0.22. When additional harvest is added to the product mix along with a nonzero value for biofuel production and use (i.e., LR, IA, DA management when biofuel and traditional wood products are produced), it reduces the relative weight of wood fuel-based emissions on the total. Hence, Et and Eh increased less than 25% within the same management strategy, but between biofuel scenarios (i.e., biofuel + wood products compared with no biofuel). Because the proportion of conifer harvest that is used for domestic fuel (3%) is significantly lower, the industrial budget for conifer forests increased twofold (up to 46%) compared with hardwood forests. Ec increased (16–144%) in all scenarios when biofuel production was increased. Although the changes observed in Et were on the order of 3–35 g C m−2 yr−1, this is equivalent to 2 × 1012 to 7 × 1012 g C yr−1 when scaled to the entire MCI forest region.
This study mainly relied on emission factors for harvest and wood, paper, and biofuel production within the MCI region. There were at most two studies, and typically only one available to derive the emission factors used in the industrial C cycle model. However, these data were often based on region-specific studies. In this study, we assumed that the harvest and transportation emissions for wood processed at the biofuel refinery were equivalent to those in regional production of dimensional lumber, OSB, and paper (Gower et al., 2006). However, the emission factor for a given product can vary by an order of magnitude when transportation distances increase beyond the region where the material is sourced (Gower et al., 2006), hence model estimates in this study are dependent on our assumptions and those of the life-cycle studies. Increasing transport distance of harvested material could have significant impacts on model outputs and simulated net C balance, either positive or negative. There are currently no major biorefineries in the region producing ethanol from forest biomass, so our estimates here rely on location within the resource base (Fig. 6) and transportation and harvest emissions equivalent to those in the traditional forest products industry. Transportation of wood fiber comprised 94% of the total CO2-eq emissions for dimensional lumber; however, this is likely to be an extreme case because the wood was harvested in northern British Columbia, Canada, and transported throughout Canada and the United States (Gower et al., 2006). Nonetheless, the study illustrates a common challenge: woody biofuel feedstock density is often greatest in regions far from the large metropolitan areas where energy demand is greatest.
Consideration of avoided emissions from biofuel use
Although this study focused on computing the whole-system carbon budget for a large forested region, not all possible emissions were considered. One of the main perceived benefits of using biofuels for transportation is the greenhouse gas emissions avoided through substitution for fossil fuel. Several life-cycle studies have reported significant decreases in GHG emissions when bio-ethanol replaces gasoline as a transportation fuel (Graham, 2003; Kemppainen & Shonnard, 2005; Kim & Dale, 2005; Bright & Stromman, 2009). Obviously, there are infinite substitution possibilities for all the products considered (i.e., wood for fossil sources in energy generation, wood building materials for concrete and steel, etc.), but here we provide some insight on the emission-related benefits in the three scenarios where excess harvest is used only for biofuel production and the other products remain unchanged from the base. Based on data from Timpe & Aulich (2005), avoided emissions (0.75 kg CO2-eq L−1) through substitution of bio-ethanol for gasoline applied to this study, resulted in avoided emissions that were 3–9% of the total industrial emissions, depending on the management scenario. Both Marland & Schlamadinger (1997) and Schlamadinger & Marland (1996) reported that the net C balance of the forest system was dependent on the conversion efficiency of harvested wood (i.e., emissions per unit C) and the fossil fuel system that was potentially replaced, as we observed in this study. Although we did not completely address changes in greenhouse gas emissions due to reduced use of fossil fuels in favor of bio-ethanol in vehicles or substitution and leakage associated with changes in forest product production or land-use change, they should be considered in future studies of the carbon cycle.
In this study we have focused on the utilization of forest biomass to produce liquid transportation fuel, however, we acknowledge that currently this may not be the most economically competitive use given the current price of gasoline and both the cost and efficiency of ethanol production from woody biomass. However, woody biomass will certainly play a role in future fuel production due to the US governments' policy (Energy Independence and Security Act of 2007) for fuel and energy production from nonfossil sources and the sustainability of forest resources vs. using agricultural land for energy production (Kauppi & Saikku, 2009; Ohlrogge et al., 2009). Electricity generation is another important potential use of woody biomass in the United States (Ohlrogge et al., 2009) that was not considered in this study. Although not widely adopted in the study region, it is possible that this may be the most efficient use for woody biomass in the future (Campbell et al., 2009; Ohlrogge et al., 2009), especially if heat generation is combined and fossil fuels are replaced (Steubing et al., 2012). Wood pellet production (used for both residential heating and electricity generation) in the region has doubled recently (Luppold et al., 2011), which suggests that this could be an important future use of woody biomass and could easily be incorporated into the model after the appropriate life-cycle studies are completed. A whole-system modeling approach, like the one outlined in this study, is a critical step toward a complete model of the forest carbon cycle in a managed forest landscape.
This research was supported by United States Department of Agriculture grant 00406117 to S. T. Gower and in part by the DOE Great Lakes Bioenergy Research Center (DOE Office of Science BER DE-FC02-07ER64494). We thank Phil Townsend, Eric Kruger, Charles (Hobie) Perry, and Chris Kucharik for critical review of this work. We also thank five anonymous reviewers whose comments helped improve this manuscript.