2.1 The Model LPJmL

We here use the well-established [Cramer et al., 1999; Gerten et al., 2004; Friend et al., 2014] Dynamic Global Vegetation Model including managed land [Bondeau et al., 2007; Text S2, Supporting Information] to simulate the growth of natural and managed vegetation—including BPs—and the associated biogeochemical processes in a single, internally consistent framework. LPJmL embraces nine plant functional types with dynamic distributions based on bioclimatic conditions and competition for light, water, and space, and 12 crop functional types as well as pastures with prescribed distributions and management [e.g., irrigation, Jägermeyr et al., 2015] with calibrated yields until 2005 [Fader et al., 2010]. Furthermore, the model captures two second-generation bioenergy functional types with prescribed distribution in scenarios (as specified in Table 1) producing yields [Beringer et al., 2011]. With daily time steps and on a 0.5° × 0.5° grid, the model evaluates the climate-dependent transient dynamics of carbon fixation, allocation, turnover and loss in vegetation growth while accounting for interactive effects of soil moisture and atmospheric CO₂ content [for a detailed description of model setup please see also Text S2 and Schaphoff et al., 2013]. Bioenergy plantations are either woody (representing the growth characteristics of temperate willows and poplars or tropical Eucalyptus) or herbaceous (imitating Miscanthus and switchgrass) with harvest cycles of 8 years or multi-annual occurrences, respectively [Beringer et al., 2011; Heck et al., 2016; Text S1]. Simulations of dedicated BPs differ from those of corresponding natural vegetation by assuming higher productivity and harvest at regular or growth-dependent intervals. Our results capture a realistic magnitude of production as verified by a comparison of the uncalibrated simulated woody and herbaceous BP productivity with observations from field data as conducted by Heck et al. [2016].

2.2 Climate and Land Use Scenarios

We investigate the ability of tCDR to counteract future emissions associated with three climate scenarios that generate specific GMT levels by the end of the century. In the first scenario, climate develops on an unabated business-as-usual pathway resulting in a GMT rise of ∼4.5°C by 2100 [Heinke et al., 2013] and a crossing of the 2°C threshold in 2053 (Figure 1a). This climate forcing is comparable to the RCP8.5 projection [Riahi et al., 2011] with 2,085 GtC cumulative anthropogenic emissions by the end of the 21st century. On the RCP8.5 trajectory, a GMT rise by 2°C above pre-industrial level is reached by 2047 [see figure, SPM.10 in Stocker et al., 2013], hence, we extended our simulations until 2106 to capture a similar timespan (e.g., 2053–2106 instead of 2047–2100). In the second scenario, emissions are partially mitigated, leading to a GMT rise of ∼2.5°C by 2100 (and ∼1.8°C by mid-century). This forcing is comparable to the RCP4.5 scenario with 1227 GtC cumulative anthropogenic emissions by 2100. It is also equivalent to the current mitigation pledges of the Paris accord [Jeffery et al., 2015] and thus still not complies with the internationally agreed objectives. The third climate scenario follows temperature and emissions projections (1.7°C and 780 GtC cumulative emissions by 2100, respectively) of the strong-mitigation, transient RCP2.6 trajectory [van Vuuren et al., 2010].

The climate forcing and CO₂ concentrations for the first two climate scenarios were taken from the MPI-ESM ECHAM5 model and found to give intermediate biosphere responses in LPJmL when compared with four additional climate models participating in the Coupled Model Intercomparison Project Phase 3 (CMIP3) and prepared for our 2.5°C and 4.5°C trajectory following Heinke et al. [2013, see Figure S3]. For RCP2.6, climate model inputs were taken from five CMIP5 models [HadGEM2-ES, MPI-ESM-MR, CanESM2, IPSL-CM5A-MR, and MIROC-MR-CHEM; Taylor et al., 2012] and bias-corrected [Watanabe et al., 2012; Heinke et al., 2013]. The sensitivity of BPs' productivity to different levels of CO₂ and climate is rather high but within the range observed in field experiments [Figure S2, Norby et al., 2005; Hickler et al., 2008].

In the first part of our analysis, we investigate the tCDR potential necessary to hold the 2°C line in the high and the partially mitigated emission scenario, respectively, using a set of land-use scenarios differing in terms of the spatial extent of BP plantations and in terms of whether currently cultivated or uncultivated areas are considered for conversion (Figure 1b, Table 1). In these scenarios, land-use patterns for crops and pastures are fixed at year 2005 levels until BPs are implemented according to the following rationale: By utilizing all natural land suitable for growing BPs (7.4 Gha globally, 100NAT) or all present-day agricultural land (4.2 Gha, 100AGR), theoretical upper limits on tCDR potentials are established. Scenarios where 25% of the most productive either natural (e.g., parts of five major biomes, 25NAT) or agricultural land grid cells (e.g., parts of cropland and pastures, 25AGR) are converted to BPs represent very extensive versions of tCDR (3.3 or 2.2 Gha). Conversions of 10% of grid cells (1.4 Gha in 10NAT and 1.1 Gha in 10AGR, respectively) are perhaps more realistic but still very ambitious tCDR scenarios. Grid cells unsuitable for agriculture, e.g., those covered by ice, snow, or desert, were excluded. However, BPs may also be realized in apparently low-productivity regions (e.g., central Australia) as long as they belong to the 25% most productive grid cells of grass- or shrubland [following the biome classification by Ostberg et al., 2013]. For this systematic analysis of potentials, we did not adhere to the land-use patterns associated with the studied RCPs [Riahi et al., 2011; Thomson et al., 2011], since we are interested in the effectiveness of BPs to balance additional emissions under different climates and CO₂ concentrations per se, i.e., based only on production potentials for the selected areas, irrespective of whether they are considered BP areas in RCP land-use scenarios (see below for our alternative scenario that allows for a feasibility analysis under an RCP2.6 land-use pattern).

Note that grid cells and plantation types are selected for conversion such that the highest global net carbon potential—that is the highest sum of land-carbon changes from land conversion plus carbon extraction by BPs—is achieved in each scenario (global distribution shown in Figure S1).

In the second part, the feasibility of a predefined RCP2.6 land-use scenario for crop, pasture, and bioenergy plantations [van Vuuren et al., 2010; http://luh.umd.edu/data.shtml] is analyzed. In this scenario, created by an Integrated Assessment Model (IAM), global population reaches 9 billion in 2050 causing cropland expansion due to increasing food demand. Growing land-use intensity allows for agriculture to concentrate in poorer world regions while the abandoned land in wealthier regions can be used for establishment of BPs. Specifically, this means that BPs are only allowed on abandoned crop and pasture land or natural, non-forested and non-protected land. This in turn causes deforestation to meet the increasing food and energy demand and to compensate for a decreasing CO₂ fertilization effect. Land-use change is therefore only demand-driven and not by climate policies. The provided single crop land distribution was transferred to the spatial patterns of 13 crop types in LPJmL in 2005 and proportionally scaled to meet required areas (as described in Boit et al., 2016; for spatial and temporal patterns see Text S3). According to this scenario, dedicated BP areas increase to 441 Mha between 2006 and 2100 and are assumed to be herbaceous, since these have higher biomass harvest potentials than woody BPs on a global scale [see Kato and Yamagata, 2014; Heck et al., 2016]. In contrast to the previous scenarios, irrigation of BPs is explicitly considered, assuming either sustainable or unlimited schemes. For the version assuming sustainable irrigation, water withdrawal is constrained by local renewable water availability from rivers, groundwater baseflow, lakes, and reservoirs. Irrigated cropland area in 2005 [Jägermeyr et al., 2015] is assumed to increase proportionally over time (reaching 362 Mha out of 5005 Mha total agricultural land in 2100), as is the irrigated fraction of BP area (reaching 40 Mha in 2100). In the (thought experiment) scenario of unlimited irrigation, any water demand is assumed to be met in one way or the other.

2.3 Calculation of Carbon Sequestration Potentials

For BPs to be an effective tCDR tool, that is realizing net negative emissions, the extracted carbon from the atmosphere needs to be permanently immobilized and excluded from the carbon cycle. We calculate the cumulative biomass harvest carbon until 2100 but employ a conversion-efficiency rule (CEff) before accounting for land-carbon changes for the overall tCDR potential. That is, we simplify the life-cycle assessment of processed carbon and carbon products by assuming that harvested biomass carbon is passed to a carbon pool with a CEff of 50%: Half of the sequestered carbon is lost during biomass harvest, its subsequent transportation, processing, and storage on longer time scales. This simplification represents a reasonable global averaging in line with the current evidence [Lenton, 2010; Powell and Lenton, 2012; Smith et al., 2013]. However, there are various schemes by which biomass carbon could be utilized: biofuels with CEff's of ∼20%, 50%, or 70% for the processes of fermentation, liquefaction or pyrolysis, respectively [Edenhofer et al., 2011]; bioenergy with carbon capture and storage (BECCS) in geological reservoirs with theoretical CEff's of 70–99% [Lenton, 2010; Edenhofer et al., 2011; Humpenöder et al., 2014]; or biochar generation, substituting fertilizers on fields, with potential CEff's of 70–90% (Lehmann, 2007; Woolf et al., 2010). Transportation may lead to losses of 2–15% (Cannell, 2003; Smeets et al., 2007). We account for this range when analyzing the tCDR potential in the RCP2.6 scenario by applying values of CEff of 50%, 75%, and 90%. Lower values of 20% or less could lead to the unfavorable outcome that land-carbon losses exceed the carbon extraction by BPs, so they are not considered here. The replacement of natural vegetation by land conversion is treated as a one-time harvest with a 50% capture rate. As CO₂ concentrations are prescribed in the model, we cannot account for substituting fossil fuel energy carriers.

We express the tCDR potentials in terms of cumulative CO₂ uptake and in terms of “years delayed” on the emission trajectories of the similar climate scenarios of the partially mitigated RCP4.5 and unabated RCP8.5 to infer whether tCDR can extend the 2100 emissions budget for a certain period of time into the future. This is achieved by subtracting accumulated tCDR potentials in 2100 from the RCP4.5 or RCP8.5 emissions budget in 2100. By decelerating the emission accumulation with tCDR without actually leaving the trajectory we can bypass the fact that in LPJmL CO₂ concentrations are not affected by tCDR, thus fossil fuel substitution and ocean and climate feedbacks cannot be accounted for.

Under RCP2.6, BPs are required to extract 160–190 GtC from the atmosphere [van Vuuren et al., 2010; Kato and Yamagata, 2014]. This should not only refer to the biomass harvest potentials alone but should again take land-carbon changes from conversion to BPs into account. Note that tCDR potentials for RCP2.6 as simulated by LPJmL depend on the specific climate forcing provided by the different climate models. Not only may precipitation patterns differ but also GMT increases may vary, mainly due to the diverse carbon cycle responses arising from different modeling strategies [Brovkin et al., 2013; Boysen et al., 2014]: BP patterns were not uniformly implemented in coupled models' land surface schemes in CMIP5; they were treated as cropland or grassland or entirely ignored, so the resulting climate patterns vary too. These differences provide an uncertainty range for the simulated tCDR potentials presented here.

2.4 Calculation of Impacts of tCDR

We provide a rough assessment of impacts of tCDR on food production (i.e., drawbacks on current production that is estimated to be ∼3,000 kcal cap⁻¹ day⁻¹ for 7 billion people based on simulated crop yields and calorie content [see Text S3 and Wirsenius, 2000]), forest extent, and the nitrogen cycle.

Based on the concept of “planetary boundaries” [Rockström et al., 2009; Steffen et al., 2015] we also quantify the impact of further reductions of natural forest cover brought about by tCDR plantations. The land-system boundary defines thresholds of remaining forest extent for three continental forest biomes (boreal, temperate, and tropical) beyond which the Earth system would enter a new state (in a possibly irreversible way). We used this approach according to the fractional forest areas provided by LPJmL, which partly differ from those used in Steffen et al. [2015]. By scaling simulated potential forest extents to those in Steffen et al. [2015], we were able to analyze the relative change of area in our scenarios and to calculate the position with respect to the planetary boundary for land-system change.

Nitrogen limitation to plant growth is not explicitly modeled in LPJmL. Therefore, we did a post hoc estimation of the nitrogen content of the harvested and removed biomass [see Boysen et al., 2016], which can be translated into the required amount of nitrogen fertilization. We do not account here for the possible increase in N₂O emissions that could arise from this increased fertilizer application or increased fossil fuel use for producing fertilizer.

3.1 Sequestration Potentials of tCDR in a Partially Mitigated and in an Unmitigated Scenario

We find that tCDR could potentially push down GMT toward the 2°C line in a partially mitigated climate scenario that would approach 2.5°C above pre-industrial level by 2100 (corresponding to ∼320 GtC emissions between mid-century and 2100, following RCP4.5, see Figure 2). This is summarized in Table 1a. However, the required BP area needs to be >1.1 Gha of the most productive agricultural land (10AGR, Figure 2f) or ∼1.5 Gha of natural land (10NAT, Figure 2c). The theoretical tCDR potential is as large as 585–1130 GtC, or 67–165 years of delay, if all agricultural or natural land can be converted (100AGR and 100NAT, respectively, Figures 2d and 2a). However, this would have most dire consequences for food production or the biosphere. In order to maintain current global land use patterns for agriculture, the majority of natural ecosystems would be eliminated in the 100NAT scenario. Moreover, the upper ceiling for tCDR potential would be lowered by up to 20% if the conversion of natural vegetation for the sake of tCDR would not make sure that half of the natural-vegetation carbon would be permanently sequestered. Alternatively, converting all cropland and pastures into tCDR area (100AGR) while safeguarding natural ecosystems, would imply that all land-based food and fiber production was abandoned. The implications still remain severe if only a quarter of natural or agricultural land is taken for BPs, thereby stretching 2100's carbon budget by 56–74 years (Figures 2e and 2b). Additionally, simply allowing re-growth of natural vegetation on the same areas as in 25AGR (25ARG_nv, Figure 2e) would reduce potentials by >40% compared to BPs with 454 GtC permanent carbon extraction by 2100.

These tCDR potentials could be increased by 10–15% if BPs were implemented earlier under the same climate conditions, e.g., when the 1.5°C warming line was reached around 2038 [Boysen et al., 2016]. However, starting BPs mid-century when CO₂ levels and additional fertilization are higher, as in the unabated climate scenario, can also be beneficial for BPs leading to similar tCDR potentials as in Boysen et al. [2016] (see supporting information [Leipprand and Gerten, 2006; Luo et al., 2008] and Figure S2). For example, the 100AGR scenario removes 649 GtC between 2038 and 2100 on a partially mitigated pathway, while the 100AGR on an unabated pathway captures 705 GtC between mid-century and 2100. However, tCDR is much less effective in postponing the pertinent carbon-budget exhaustion: The maximum achievable tCDR potential translates into a delay by 28–67 years in the 100AGR and 100NAT scenarios, respectively. The more conservative but still extensive scenarios (10AGR and 10NAT) can only provide 13–16 years of delay—on average three times less than in the partially mitigated scenario. This shows that even though the CO₂ “fertilization” effect on plant growth is about 17–20% higher on the unabated mitigated pathway, strong emissions abatement efforts are necessary as well as carbon extraction technologies such as tCDR. Staying on the unabated pathway would mean not only a failure of mitigation in the first place, but would also predetermine tCDR to fail its ultimate objective!

These results allow us to investigate the upper theoretical ceilings of tCDR following systematic land conversion scenarios by minimizing the uncertainties from the development of complex, actual land-use patterns (as done in the RCP2.6 study below). However, they also show that “realistic” tCDR potentials, following multi-dimensional optimal pathways of land transformation are likely to remain far below scenarios such as the 100%, 25%, or 10% scenarios used in this study.

3.2 Side-Effects of Large-Scale Biomass Plantations

Irrespective of the underlying emissions scenario, large-scale tCDR deployment would be associated with impacts that are likely to be ecologically intolerable and socially unacceptable (Table 1b). The land conversion towards tCDR following 25NAT implies widespread loss of habitats, thus further reducing biodiversity and modifying ecosystems which are already under pressure [Ostberg et al., 2015] and face severe risks of change under anthropogenic global warming [Ostberg et al., 2013; Warszawski et al., 2014]. The global forest extent, currently estimated to consist of 62% natural sub-systems [Steffen et al., 2015], would be halved in this tCDR scenario. When converting “only” 10% of natural land, still almost 1.4 Gha of habitats would be lost or degraded—an area corresponding to half of today's pasture extent.

From calculating the nitrogen content in the globally harvested biomass under unabated conditions, we find that this harvesting would extract 96–151 TgN yr^–1 on 10–25% of the agricultural area (in addition to the demand on the remaining cropland). This is of a magnitude comparable to today's worldwide nitrogen demand of 147 TgN yr^–1 in 2014 [Food & Agriculture Organization of United Nations, 2015; Steffen et al., 2015], which already has led to transgression of the suggested planetary boundary for nitrogen by a factor of two [Steffen et al., 2015; Food and Agricultural Organisation United Nations, .]. As another consequence, substantial extra amounts of non-CO₂ greenhouse gases would be released into the atmosphere after fertilizer application or during the process of fertilizer generation. Thus, our simulated tCDR potentials may be overestimated since we neither account for nutrient limitation of plant growth (nitrogen and phosphorus), nor for the emissions associated with producing and applying fertilizers.

Agricultural calorie production on cropland would be reduced by 43–73% when converting the most suitable 10–25% of cropland for the purpose of tCDR. In view of a world inhabited by at least 9 billion people in 2050, it is unlikely that such deficits could be overcome by sheer management intensification or improvement [Bajželj et al., 2014]. Transforming merely all pastures (i.e., keeping all cropland while eliminating the entire production of meat and dairy products, 100AGR_p) would not result in substantial climate benefits: While pastures are more extensive than croplands, they are also less productive for BP plantations.

Further impacts could arise from generating biogeophysical and biogeochemical effects through the land conversion towards large-scale BPs [Pongratz et al., 2011; Arora and Montenegro, 2011; Brovkin et al., 2013]. These effects could include reductions in surface albedo and alterations of moisture fluxes [see Text S1 and Boysen et al., 2016] or additional greenhouse gas emissions from fertilizer applications. Evidently, fully coupled simulations are needed to assess these climate feedbacks (including changing atmospheric CO₂ concentrations and ocean responses [Zickfeld et al., 2013; Tokarska and Zickfeld, 2015], something that currently cannot be accomplished because most Earth system models still lack a process-based implementation of BPs in the way LPJmL does.

3.3 Carbon Sequestration Potentials of tCDR in a Transient Mitigation Scenario (RCP2.6)

Harvested biomass carbon on 441 Mha BPs accumulates on climate-model average 152, 325, and 449 GtC under rain-fed, sustainable, and unlimited irrigation conditions, respectively (Table S2). While these results are uniform across the models, the same is not true for land-carbon changes after the conversion of original land cover to BPs, i.e., the resulting tCDR potential by 2100 varies strongly (Table 2). Applying a conversion efficiency of 50% (as before) delivers for instance ranges of 10–64 GtC (mean 57 GtC) net carbon sequestration for non-irrigated BPs (Figure 3a). The reason is that the prescribed climate forcing retrieved from each single climate model might cause unfavorable growing conditions for BPs in some cases. If CEff was not increased, unlimited irrigation would be necessary to achieve the required tCDR potential of 160–190 GtC in 2100. An increase of CEff to 75% in combination with sustainable irrigation on selected areas (40 Mha) could produce the desired tCDR outcome though. However, such an increase in the sustainably irrigated area would in turn put substantial pressure on global water use. In fact, the water consumption on dedicated BP land would increase the total water consumption on agricultural land [∼1,000 km³ yr⁻¹ in 2090–2100) by 13% on model average (see Table S3, comparable to Heck et al., 2016]. This would strongly increase by a factor of 20 (up to 2,500 km³ yr⁻¹) if all BPs were equipped with unlimited water supply.

Remarkably, increasing CEff to 90% without any irrigation would still not be sufficient to get close to the desired tCDR amount. This means that very strong efforts regarding water (and fertilizer) management, infrastructure, and technical development would be globally necessary to fulfill the promises of this widely used scenario for successful mitigation. Our result agrees with a previous study claiming that highly productive BPs, supported by heavy irrigation and fertilization, would be needed in the RCP2.6 narrative [Kato and Yamagata, 2014]. The implications would again include biogeophysical and biogeochemical effects, which were not considered in the land-selection process of the underlying IAM, while food production would need to increase by ∼20% until 2050 compared to 2005 levels in our model. Thus, tCDR potentials on the RCP2.6 pathway are found to be smaller than anticipated. This finding reveals significant trade-offs when trying to reach the total tCDR volume necessary for holding the 2°C line—even more so, as that scenario was meant to “start” about one decade ago! Hence, assumptions made by IAMs need to be carefully assessed regarding the true abilities of the biosphere, which are independent of any socio-economic considerations.