Investigating afforestation and bioenergy CCS as climate change mitigation strategies

MAgPIE is a spatially explicit, global land-use allocation model and projects land-use dynamics in ten-year time steps until 2095 using recursive dynamic optimization (Lotze-Campen et al 2008, Popp et al 2010). The objective function of MAgPIE is the fulfilment of food, livestock and material demand at minimum costs under socio-economic and biophysical constraints. Demand is based on exogenous future population and income projections (see scenario section), while price-induced changes in consumption are not reflected. Major cost types in MAgPIE are: factor requirement costs (capital, labor and fertilizer), land conversion costs, transportation costs to the closest market, costs for R&D (technological change) and costs for GHG emission rights. For long term investments, like land conversion or R&D, we assume a time horizon of 30 years and an annual discount rate of 7%, which reflects the opportunity costs of capital at the global level (IPCC 2007, chapter 2.4.2.1). While socio-economic constraints like trade liberalization and forest protection are defined at the regional level (ten world regions) (figure S1), biophysical constraints such as crop yields, carbon density and water availability, derived from the global hydrology and vegetation model LPJmL (Bondeau et al 2007, Müller and Robertson 2014), as well as land availability (Krause et al 2013), are introduced at the grid cell level (0.5 degree longitude/latitude; 59 199 grid cells). Due to computational constraints, all model inputs in 0.5 degree resolution are aggregated to 500 clusters for the optimization process based on a k-means clustering algorithm (Dietrich et al 2013). During the optimization process, the cluster level is the finest resolution. The clustering algorithm combines grid cells to clusters based on the similarity of data for each of the ten world regions. If, for instance, two grid cells with similar land patterns are merged into one cluster, the algorithm preserves the land area available by summing up the area of the two grid cells. Investment in R&D in the agricultural sector translating into yield-increasing technological change is a variable in MAgPIE and can therefore endogenously enhance food and bioenergy crop yields (Dietrich et al 2014). Finally, the cost minimization problem is solved through endogenous variation of spatial production patterns, land expansion (both at the cluster level) and yield-increasing technological change (at the regional level) (Lotze-Campen et al 2010).

MAgPIE features land-use competition based on cost-effectiveness at cluster level among the land-use related activities food, livestock and bioenergy production as well as afforestation. Available land types are cropland, pasture, forest and other land (e.g. non-forest natural vegetation, abandoned agricultural land, desert). The forestry sector, in contrast to the agricultural and livestock sector, is currently not implemented dynamically in MAgPIE. Therefore, timberland needed for wood production, consisting of forest plantations and modified natural forest, is excluded from the optimization, which is about 30% of the initial global forest area (4235 mio ha). In addition, other parts of forestland, mainly undisturbed natural forest, are within protected forest areas, which is about 12.5% of the initial global forest area (FAO 2010). Altogether, about 86% of the world's land surface is freely available in the optimization of the initial time-step.

MAgPIE calculates emissions of the Kyoto GHGs carbon dioxide (CO₂), nitrous oxide (N₂O), and methane (CH₄) (Bodirsky et al 2012, Popp et al 2010, 2012). Mitigation of GHG emissions is stimulated by an exogenous tax regime on GHG emissions (see scenario section). The GHG tax is multiplied with GHG emissions in order to calculate GHG emission costs, which enter the cost minimizing objective function of MAgPIE. For instance, CO₂ emissions from land-use change can be reduced through avoided deforestation (carbon stock conservation). But unlike N₂O and CH₄ land-use emissions, CO₂ emission from the land-use system can become negative if photosynthetic carbon uptake from the atmosphere outweighs carbon release to the atmosphere from plant decomposition and land-use change (carbon stock increase). Therefore, land-based climate change mitigation via afforestation or bioenergy CCS is incentivized by the revenue from the GHG tax regime for carbon removal from the atmosphere. A detailed description of the underlying formulas is available in the supplementary data, available at stacks.iop.org/ERL/0/000000/mmedia. For the conversion of N₂O and CH₄ emissions into CO_2eq we use GWP100 (IPCC (2013)).

In MAgPIE, dedicated lignocellulosic biomass (rainfed only) can be converted to secondary energy via different conversion routes. The carbon released from biomass during combustion is captured and stored underground using CCS technology.

Herbaceous and woody bioenergy yields at grid cell level for the initialization of MAgPIE are derived from LPJmL (Beringer et al 2011). While LPJmL simulations supply data on potential yields, i.e. yields achieved under the best currently available management options, MAgPIE aims to represent actual yields. Therefore, LPJmL yields are reduced using information about observed land-use intensity (Dietrich et al 2012) and agricultural area (FAO 2013). For instance, in China (CPA) herbaceous bioenergy yields from LPJmL are reduced by about 45% to obtain actual yields for the initialization of MAgPIE (table 1). MAgPIE bioenergy yields can exceed LPJmL bioenergy yields over time as endogenous investments in R&D push the technology frontier. Higher bioenergy yields are associated with increased carbon uptake from the atmosphere per unit area.

Bioenergy CCS can provide energy and remove carbon from the atmosphere at the same time. Due to this versatility, bioenergy CCS is an attractive mitigation option in scenarios with ambitious climate targets. The largest share of profits in these scenarios comes from the carbon sequestration and not from the energy portion of the bioenergy CCS technology (Rose et al 2014). In this study, we focus on the carbon removal potential of bioenergy CCS and therefore disregard the value of the energy produced. We implemented three different conversion routes with CCS technology in the model: biomass to hydrogen (B2H2), biomass integrated gasification combined cycle and biomass to liquid (B2L) (Klein et al 2014). Levelized costs of energy (LCOEs) are calculated using the time horizon of 30 years and the discount rate of 7% (see MAgPIE description), which are both within the range of common assumptions for LCOE calculations (IEA and OECD NEA 2010, chap 2.3). The LCOEs include initial investment costs in infrastructure as well as operational and maintenance costs. The B2H2 technology in combination with CCS features a higher conversion efficiency (55%) and carbon capture rate (90%) at lower LCOEs (8 $ GJ⁻¹) than the other available technologies (based on Klein et al 2014). Demand for bioenergy in this study does not rely on exogenous trajectories, but is derived endogenously as a response to the GHG tax, which rewards carbon removal due bioenergy CCS, while the value of energy produced due to bioenergy CCS is disregarded. The cost minimizing objective function of MAgPIE in combination with carbon removal being the only incentive in the model to employ bioenergy CCS renders the B2H2 technology superior to the other available conversion routes. The geological carbon storage capacity is constrained at the regional level (table S3), which adds up to 3960 GtCO₂ at the global level (Bradshaw et al 2007). We assume a lifetime of the CCS technology of 200 years (Szulczewski et al 2012) and therefore limit the annual geological injection of carbon to 0.5% yr^–1 in terms of the geological carbon storage capacity, which results in an annual realizable geological injection rate of about 20 GtCO₂ yr^–1 globally. As biomass can be traded, the location of geological carbon storage can differ from the location of biomass production. Levelized costs for transportation and injection of captured carbon are at 9 $/tCO₂ (Klein et al 2014).

Compared to bioenergy CCS, afforestation can be considered as low-tech land-based mitigation strategy, since no technical infrastructure for processing is needed. In MAgPIE, afforestation is a managed regrowth of natural vegetation. The regrowth is managed in that way as endemic seed sources are put in place manually as part of the land conversion process. Regrowth of natural vegetation affects vegetation, litter and soil carbon stocks, which are calculated as the product of carbon density and afforestation area (see online supplementary data for details). Vegetation, litter and soil carbon density of potential natural vegetation in 1995 at grid cell level is derived from LPJmL (figure S4). Vegetation carbon density increases over time along S-shaped growth curves (figure 1). The vegetation carbon density growth curves are based on a Chapman–Richards volume growth model (Murray and von Gadow 1993, Gadow and Hui 2001), which is parameterized using vegetation carbon density of potential natural vegetation (figure S4(a)) and climate region specific mean annual increment (MAI) and MAI culmination age (IPCC 2006). Litter and soil carbon density are assumed to increase linearly towards the value of potential natural vegetation (figures S4(b) and (c)) within a 20 years time frame (IPCC 2000). The initial value for vegetation and litter carbon density is assumed to be 0, while the initial grid-cell specific value for soil carbon density is the average between cropland and potential natural vegetation soil carbon density.

Zoom In Zoom Out Reset image size

In MAgPIE, the decision to invest in afforestation depends on the benefit-cost ratio over the time horizon of 30 years, which is a common crediting period for afforestation projects (United Nations 2013). Firstly, cumulative carbon uptake over the time horizon is calculated as the product of new afforestation area and carbon density (vegetation, litter and soil) at age 30. Secondly, the benefit of an afforestation activity is calculated as the product of this cumulative carbon uptake 30 years ahead and the level of the current GHG tax (see online supplementary data for more details). Finally, for comparability with the annual bioenergy CCS activity, this future cash flow is annualized using the discount rate of 7% (see MAgPIE description). Land conversion and management costs are based on Sathaye et al (2005). Regional costs for the conversion of any land type into afforestation land range between 849 $ ha⁻¹ (SAS) and 2484 $ ha⁻¹ (NAM) (table S2) in the initial time step and are proportionally scaled with GDP for future time steps. Total land conversion costs are annualized. Annual management costs range between 2 $ ha^–1 yr^–1 (FSU) and 127 $ ha^–1 yr^–1 (AFR). Contrary to bioenergy CCS, technological change has no direct effect on the carbon removal potential of afforestation.

Our scenarios are based on the shared socio-economic pathways (SSPs) for climate change research (O'Neill et al 2014). It should be noted that the SSPs do not incorporate climate mitigation policies by definition. In this study, we choose SSP 2, a 'Middle of the Road' scenario with intermediate socio-economic challenges for adaptation and mitigation. Food, livestock and material demand (figure S2) is calculated using the methodology described in Bodirsky et al and the SSP 2 population and GDP projections (IIASA 2013).

We assume a GHG tax (Tax30) on Kyoto gases (CO₂, N₂O, CH₄) that increases nonlinearly at a rate of 5% yr^–1 (Kriegler et al 2013). The GHG tax has a level of 30 $/tCO_2eq in 2020 and starts in 2015 (figure S3). The resulting GHG tax with prices of 102 $/tCO_2eq in 2045 and 1165 $/tCO_2eq in 2095 is close to GHG price trajectories required to limit global average temperature increase to 2 °C above pre-industrial levels with a probability of 50% (Rogelj et al 2013b). Due to this ambitious climate change mitigation target, biophysical climate impacts on crop yields and carbon densities are assumed to be weak and are therefore not regarded in this study.

We investigate four scenarios, which cover two dimensions: GHG tax and availability of carbon removal options (table 2). In the business as usual scenario (BAU), no tax on GHG emissions is applied, i.e. there is no incentive to avoid GHG emissions. In the mitigation scenarios, the GHG tax penalizes all positive GHG emissions from the land-use system and rewards negative CO₂ emissions from afforestation in AFF, from bioenergy CCS in BECCS, and from both in AFF+BECCS.

To address the uncertainty in exogenous model parameters, we investigate the sensitivity of our simulations to changes in model parameterization. Crucial parameters in the context of this study are the geological carbon storage capacity (CCS capacity), the GHG tax, the time horizon for investment decisions, the discount rate, and assumptions about future bioenergy yields. Table 3 summarizes the parameter settings for the sensitivity analysis.

CCS capacity: Bradshaw et al (2007) estimates a range for geological carbon storage capacity at the global level of 100–200 000 GtCO₂. For DEFAULT, we use 3960 GtCO₂. For the sensitivity analysis we vary this value by factor 20 in both directions (198 GtCO₂ in LOW, 79 200 GtCO₂ in HIGH). Based on Szulczewski et al (2012), we assume a lifetime of CCS of 200 years. Therefore, we limit the annual injection of carbon to 0.5% yr^–1 in terms of the total CCS capacity. For DEFAULT this results in 20 GtCO₂ of the total CCS capacity, for LOW in 1 GtCO₂ yr^–1 and for HIGH in 396 GtCO₂ of the total CCS capacity globally.

GHG tax: in DEFAULT, the GHG tax has a level of 30 $/tCO_2eq in 2020, starts in 2015 and increases by 5% yr^–1. For LOW and HIGH, the level of the GHG tax is 5 and 50 $/tCO_2eq in 2020 respectively. The range for the sensitivity analysis is based on Kriegler et al (2013).

Time horizon: in DEFAULT, the time horizon for investments is 30 years, which is common in the energy sector as well as for afforestation projects. For LOW and HIGH we chose 10 and 50 years respectively (IEA and OECD NEA 2010, United Nations 2013).

Discount rate: in DEFAULT, the annual discount rate is 7%, which reflects the opportunity costs of capital at the global level. For the sensitivity analysis, we vary the discount rate by 3% points in both directions, based on a literature range of 4–12% (IPCC (2007), chap 2.4.2.1).

Bioenergy yields: in DEFAULT, bioenergy yields are variable and assumed to increase, along with food crop yields, due to endogenous technological change in the agricultural sector. In LOW, technological change in the agricultural sector has no impact on bioenergy crop yields (i.e. bioenergy yields are fixed at the initial level), while food crop yields can still increase due to technological change.

To investigate the role of the algorithm used for clustering our high resolution input data, we perform sensitivity analysis with different numbers of spatial cluster units (100–500) in addition.

In 1995, total land cover (12 907 mio ha) consists of food crop (1425 mio ha), pasture (3073 mio ha), forest (4235 mio ha) and other land (4174 mio ha) (figure 2). In the BAU scenario (no GHG tax), food crop area increases by about 300 mio ha until 2095, mainly at the expense of forestland. In the second half of the century, pasture area decreases due to stabilizing livestock demand (figure S2) in combination with average yield increases of about 0.48% yr^–1 (figure S10), leading to an increase of abandoned agricultural land. In the mitigation scenarios, the GHG tax on land-use change emissions keeps forestland almost constant over time. Afforestation emerges as cost-efficient mitigation strategy from 2015 (start of GHG tax at 24 $/tCO_2eq) and increases, mainly at the expense of pasture and food crop area, to 2773 mio ha in AFF until 2095. Endogenous yield increases, accompanied by changes in spatial production patterns, compensate for the reduced agricultural area. In AFF, the cost–efficient level of average yield increases in the agricultural sector is 1.21% yr^–1 throughout the 21st century (figure S10). Bioenergy CCS comes into play much later than afforestation, as it is cost-efficient first in 2065 (270 $/tCO_2eq). Bioenergy area increases to 508 mio ha until 2095 in BECCS, mainly at the expense of food crop area. Total dedicated bioenergy production, mainly herbaceous crops, stabilizes at 237 EJ yr^–1 until 2095 (figures S6, S7). In the combined setting, AFF+BECCS, afforestation area (2566 mio ha) is slightly smaller compared to AFF, while bioenergy area (300 mio ha) is almost halved compared to BECCS. Despite the smaller bioenergy area, bioenergy production remains at 237 EJ yr^–1 in 2095 in AFF+BECCS, which is reflected in a higher level of average yield increases in AFF+BECCS (1.37% yr^–1) compared BECCS (1% yr^–1) (figure S10).

Zoom In Zoom Out Reset image size

The maps in figure 3 illustrate the spatial distribution of afforestation and bioenergy area for the standalone scenarios (AFF/BECCS) and the combined setting (AFF+BECCS) in 2095. In the standalone scenarios, afforestation area is found in many world regions, predominantly in Sub-Sahara African, Latin America, China, Europe and the USA, while bioenergy area is mostly found in the northern hemisphere in the USA, China and Europe. In the combined scenario, afforestation area is similar to AFF. But due to competition for land between the two carbon removal options, afforestation area is reduced in favor of bioenergy area in the USA and China in AFF+BECCS. There are several reasons why the USA, China and Europe are the main bioenergy producers. We provide insight in subsection 'Bioenergy CCS and the role of yield increases' along with figure 5.

Zoom In Zoom Out Reset image size

In the BAU scenario, CO₂ emissions from the land-use system accumulate to 177 GtCO₂ until 2095 (figure 4). The peak in mid-century is mainly caused by deforestation, while the following decline in CO₂ emissions is due to ecological succession on abandoned agricultural land. In the mitigation scenarios, the described land-use dynamics lead to net carbon removal from the atmosphere. More precisely, carbon is detracted from the atmosphere by photosynthesis and is either biologically sequestered via afforestation or geologically sequestered via bioenergy CCS. In AFF, land conversion into afforestation area increases cumulative CO₂ emissions in 2015, followed by continuous carbon removal of about 10 GtCO₂ yr^–1 throughout the 21st century. Until 2095, carbon removal in AFF accumulates to 703 GtCO₂. In BECCS, cumulative CO₂ emissions are almost constant until bioenergy CCS becomes cost-efficient as mitigation strategy in 2065 at GHG prices of 270 $/tCO_2eq. From 2065, carbon removal in BECCS is about 20 GtCO₂ yr^–1, which cumulates to 591 GtCO₂ until 2095. In AFF+BECCS, carbon dynamics are similar to AFF until bioenergy CCS becomes competitive as mitigation option in addition to afforestation in 2055. Carbon removal in AFF+BECCS is about 25 GtCO₂ yr^–1 in from 2065 to 2095, which results in cumulative carbon removal of 1000 GtCO₂ until 2095. In 2095 in BECCS and AFF+BECCS, the constraint on the annual geological carbon injection rate is binding (20 GtCO₂ yr^–1), while cumulative carbon storage capacity (3960 GtCO₂) would last for approximately another 150 years.

Zoom In Zoom Out Reset image size

Contrary to afforestation, the carbon removal rates per unit area of bioenergy CCS can be enhanced through yield increases. Figure 5 illustrates the potential annual carbon removal rates during the optimization for 1995 and 2095 in the AFF+BECCS scenario, i.e. the annual carbon removal rates shown here represent realizable, but not necessarily realized, carbon removal rates (compare to figure 3). In the remainder of this subsection, we talk about annual realizable carbon removal rates. In 1995, afforestation shows higher carbon removal rates than bioenergy CCS in the majority of cells, with highest carbon removal rates in the tropics (about 20 tC ha^–1 yr^–1). In 2095, the picture is fundamentally different due to yield-increasing technological change on bioenergy crops, which increases carbon removal per unit area. In AFF+BECCS, average yield-increase throughout the century is at 1.38% yr^–1 (figure S10), which more than triples initial bioenergy yields until 2095 (figure S5). By end-of-century bioenergy CCS exceeds the carbon removal rates of afforestation in the tropics (about 25–30 tC ha^–1 yr^–1). However, bioenergy production does not take place in the tropics but mainly in the USA, China and Europe (figure 3) for three reasons. First of all, the USA and China exhibit higher carbon removal rates in 2095 (about 30–40 tC ha^–1 yr^–1) compared to the tropics. Second, the tropical regions are the most attractive places for afforestation. Third, bioenergy production relies on transport infrastructure, which is much more sophisticated in Europe, USA and China than in the tropics (Nelson 2008). Bioenergy yield gains go along with increased fertilizer use, which drives N₂O emissions. In 2095, cumulative N₂O emission in BECCS and AFF+BECCS are about 30–50 GtCO_2eq higher compared to BAU or AFF (figure 4), although N₂O emissions are penalized by the GHG tax.

Zoom In Zoom Out Reset image size

In order to test the stability of our results, we perform sensitivity analyses with crucial exogenous parameters (table 3). Figure 6 shows the results in terms of land and carbon dynamics at the global level. Regional results can be found in figure S8.

Zoom In Zoom Out Reset image size

The constraint on the annual geological carbon injection rate is crucial for the scenarios with bioenergy CCS. With 1 GtCO₂ yr^–1 and 20 GtCO₂ yr^–1 the constraint is binding, which indicates that the mitigation potential of bioenergy CCS is mostly limited by the annual geological carbon injection rate. However, with a potential of 396 GtCO₂ yr^–1 the constraint is not binding, which indicates that the potential of bioenergy CCS is also limited by other factors like land availability and costs associated with bioenergy production. Bioenergy production is 530 EJ yr^–1 in HIGH, compared to 237 EJ yr^–1 in DEFAULT and 4 EJ yr^–1 in LOW (figure S7). In the combined setting, AFF+BECCS, land demand is similar for all parameter settings, while the difference in carbon removal is about 500 GtCO₂. This can be explained by considering that in the combined setting in HIGH average annual yield increases are at 1.5% yr^–1 compared to 1.25% yr^–1 in LOW (figure S10).

The carbon removal potential is highly sensitive to different levels of the GHG tax, which is the only driver for land-based mitigation in this study. In general, different GHG tax trajectories influence the point in time when bioenergy CCS and afforestation are cost-efficient, which translates into different mitigation potentials in 2095. While bioenergy CCS is cost-efficient starting from carbon prices of 165 $/tCO_2eq, afforestation emerges as cost-efficient at prices of 6 $/tCO_2eq. Therefore, the impact of changes in the GHG tax trajectory on the mitigation potential is higher in scenarios with bioenergy CCS. In AFF+BECCS, the range of sensitivity for the mitigation potential is about 900 GtCO₂. In general, the degree of sensitivity decreases with higher GHG tax levels, especially for afforestation.

Lower annual discount rates (4%) mostly affect the carbon removal potential of bioenergy CCS as lower discount rates facilitate long term investments in R&D translating into agricultural yield increases. On the contrary, higher discount rates (10%) increase the charges for credit, which is reflected in average annual technological change rates of 1.25% yr^–1 in HIGH and 1.45% yr^–1 in LOW (figure S10). The range of sensitivity for the mitigation potential is about 200 GtCO₂ for BECCS and 300 GtCO₂ for AFF+BECCS.

In terms of land, the time horizon for investment decisions mostly affects afforestation. With a time horizon of ten years, afforestation area accumulates to about 1500 mio ha, while with a time horizon of 30 or 50 years afforestation area is about 3000 mio ha, which translates into a difference in carbon removal of about 300 GtCO₂. The sensitivity of afforestation to the time horizon can be explained by recalling the shape of the forest growth curves in figure 3. The mitigation potential of bioenergy CCS is also affected as a shorter lifetime of investments in CCS infrastructure increases the costs associated with bioenergy CCS.

When bioenergy yields are fixed at their initial level, bioenergy CCS is less attractive as mitigation strategy. In BECCS, bioenergy production is reduced to 74 EJ yr^–1 in LOW compared to 237 EJ yr^–1 in DEFAULT, which results in a reduction of the mitigation potential of about 500 GtCO₂ until 2095. In the combined setting, AFF+BECCS, bioenergy CCS is no longer competitive with afforestation when bioenergy yield are not allowed to increase in the future, which reduces the mitigation potential in LOW compared to DEFAULT by about 300 GtCO₂.

The range of sensitivity across different numbers of spatial cluster units (100–500) during the optimization is small for AFF and BECCS (∼50 GtCO₂), while it is more pronounced in the combined setting (∼200 GtCO₂). In general, we observe a small trend towards less carbon removal with a higher number of cluster units (figure S12).