Biomass production in plantations: Land constraints increase dependency on irrigation water

Integrated assessment model scenarios project rising deployment of biomass‐using energy systems in climate change mitigation scenarios. But there is concern that bioenergy deployment will increase competition for land and water resources and obstruct objectives such as nature protection, the preservation of carbon‐rich ecosystems, and food security. To study the relative importance of water and land availability as biophysical constraints to bioenergy deployment at a global scale, we use a process‐detailed, spatially explicit biosphere model to simulate rain‐fed and irrigated biomass plantation supply along with the corresponding water consumption for different scenarios concerning availability of land and water resources. We find that global plantation supplies are mainly limited by land availability and only secondarily by freshwater availability. As a theoretical upper limit, if all suitable lands on Earth, besides land currently used in agriculture, were available for bioenergy plantations (“Food first” scenario), total plantation supply would be in the range 2,010–2,300 EJ/year depending on water availability and use. Excluding all currently protected areas reduces the supply by 60%. Excluding also areas where conversion to biomass plantations causes carbon emissions that might be considered unacceptably high will reduce the total plantation supply further. For example, excluding all areas where soil and vegetation carbon stocks exceed 150 tC/ha (“Carbon threshold savanna” scenario) reduces the supply to 170–290 EJ/year. With decreasing land availability, the amount of water available for irrigation becomes vitally important. In the least restrictive land availability scenario (“Food first”), up to 77% of global plantation biomass supply is obtained without additional irrigation. This share is reduced to 31% for the most restrictive “Carbon threshold savanna” scenario. The results highlight the critical—and geographically varying—importance of co‐managing land and water resources if substantial contributions of bioenergy are to be reached in mitigation portfolios.


| INTRODUCTION
Visions on a circular and bio-based economy have been formulated in response to concerns about resource scarcity and impacts associated with unsustainable use of renewable and nonrenewable resources (Geissdoerfer, Savaget, Bocken, & Hultink, 2017;German Bioeconomy Council, 2018;Ghisellini, Cialani, & Ulgiati, 2016;Gregson, Crang, Fuller, & Holmes, 2015;Hetemäki et al., 2017;Hobson, 2016;Hobson & Lynch, 2016;Priefer, Jörissen, & Frör, 2017). Biomass is increasingly used to displace nonrenewable resources (especially fossil fuels) in response to energy and climate policies (Guo, Song, & Buhain, 2015). As biomass resources are of major significance for the economy, business, and industry in many countries, biomass use also increases in response to policies intending to promote innovation and growth in industries toward a more sustainable society. Bioenergy is often assigned an important role in the future energy mix and is proposed to contribute to improved energy security, climate change mitigation, rural development, and other social and economic objectives (Souza, Victoria, Joly, & Verdade, 2015). A review by the Intergovernmental Panel on Climate Change (IPCC) of 164 long-term energy scenarios showed projected bioenergy deployment levels in year 2050 at 80-150 and 118-190 EJ per year for 440-600 and <440 ppm CO 2 eq concentration targets, respectively (25th and 75th percentiles) (Edenhofer et al., 2011). IPCC WG3 reported in the Fifth Assessment Report that widespread deployment of bioenergy with CO 2 capture and storage (BECCS) in climate stabilization scenarios indicate that this option can become important if the needed technologies and resources are available (Edenhofer et al., 2014). However, BECCS is yet unproven as a climate change mitigation option and researchers caution that the possibility of achieving negative emissions in the future may become a dangerous distraction from near term emission abatement efforts (Azar, Johansson, & Mattsson, 2013;Fuss et al., 2014).
The possible future size of biomass resources is a subject that divides researchers. Estimates vary widely due to differences in the approaches used to consider important factors, which in themselves are uncertain. Population development (Bodirsky et al., 2015;Lutz, 2013), the evolution of consumer behavior, for example, diet (Wirsenius, Azar, & Berndes, 2010), and economic and technological development (Azar, Lindgren, Larson, & Möllersten, 2006) together determine future biomass demands for food and other biobased products. Supply side factors such as crop yields, water use efficiency, and adaptation to specific growing conditions (Beddington et al., 2012;Müller et al., 2015;Neumann, Verburg, Stehfest, & Müller, 2010) determine how this biomass demand in turn translates into demands for land, water, and other resources. Supply side factors also influence prospects for bioenergy. Studies may in different ways include restrictions on access to land, water, and other resources, which are intended to reflect societal priorities, for example, related to nature protection and scarcity/vulnerability of resources (Beringer, Lucht, & Schaphoff, 2011). For recent overviews focusing on bioenergy resources, see Slade, Bauen, and Gross (2014;Creutzig et al., (2015).
Despite the uncertainties, it can be concluded that organic postconsumer waste, and residues and by-products from the agricultural and forest sectors, can make important contributions. But these biomass sources will not suffice to meet the levels of biomass demand for energy found in many climate stabilization scenarios (Edenhofer et al., 2011(Edenhofer et al., , 2014. Dedicated biomass production systems (henceforth designated "biomass plantations") would be needed for meeting part of the biomass demand in such scenarios.
This study provides a comprehensive model-based quantification of global annual biomass supply from biomass plantations (henceforth designated "plantation supply") while explicitly accounting for regional biophysical climatic and hydrological conditions. We pay particular attention to the question how different rates of irrigations, different irrigation efficiencies, and ecological constraints to freshwater use affect the plantation supply. We also include detailed and spatially explicit datasets of potential land access restrictions to assess whether land or water is the predominant limiting factor regionally and globally. Regional climatic and hydrological conditions, and land/water use constraints, have been considered in earlier studies (Beringer et al., 2011;Berndes, 2002;Bonsch et al., 2016;Dornburg et al., 2010;Fraiture, Giordano, & Liao, 2008;Gheewala, Berndes, & Jewitt, 2011;Jackson et al., 2005;King et al., 2013;Smith et al., 2016) but, to the best of our knowledge, the interplay between land and water availability has not been addressed at the level of detail captured in this study. Water availability and use is determined at river basin scale to ensure that the interaction between upstream and downstream water availability and use is considered. Economic considerations are not included, and the availability of land and water resources for bioenergy is set exogenously to reflect competing biomass demand and nature/resource protection requirements. As such, our scenarios should be understood as indicative of how different degrees of access to, and management of, land and water resources influence the plantation supply achievable at global and regional level. available for biomass plantations; (b) the availability of (irrigation) water; and (c) the land and water productivity of the biomass production systems. All these features are simulated by the global dynamic vegetation model LPJmL, here operated at a 0.5°spatial resolution and daily time steps. LPJmL simulates the growth dynamics of natural and agricultural vegetation depending on daily meteorological conditions and soil characteristics and was used to assess potential yields of energy crops (Bondeau et al., 2007;Lapola et al., 2010;Sitch et al., 2003). It considers nine plant functional types and 12 crop functional types, respectively, and three bioenergy plantation systems, one grass type and two short-rotation coppice (SRC) tree types (Beringer et al., 2011). The model calculates closed balances of carbon fluxes and pools and water fluxes including river routing and irrigation water use (Jägermeyr et al., 2015;Rost et al., 2008). Photosynthesis is simulated following the Farquhar model approach (Sitch et al., 2003). Carbon and water dynamics are closely linked so that the effects of changing climate and water availability are accounted for (Gerten, Schaphoff, & Lucht, 2007;Gerten, Schaphoff, Haberlandt, Lucht, & Sitch, 2004). Physiological and structural plant traits determine water requirements and consumption. The model's integrated approach of coupling water balance to vegetation allows to divide total water consumption into productive (transpiration) and unproductive (interception loss, soil evaporation) parts. Furthermore, water fluxes are distinguished by the contribution of precipitation (green) and irrigation (blue) water (Jägermeyr et al., 2015(Jägermeyr et al., , 2016Rost et al., 2008). The suitability of the LPJmL framework for vegetation and water studies has been demonstrated, among others by validating phenology (Bondeau et al., 2007), river discharge (Biemans et al., 2009;Gerten et al., 2004), evapotranspiration (Gerten et al., 2004;Sitch et al., 2003), and energy crop yields (Beringer et al., 2011;) (see Schaphoff et al. (2017) for an overall model evaluation). Simulations of dedicated biomass producing plantations differ from those of corresponding natural vegetation by assuming higher productivity and harvest at regular or growth-dependent intervals. Harvest management of woody and herbaceous BFTs represents reported agricultural practices on second-generation biomass plantations. Woody biomass plantations are represented by the characteristics of poplar and willows for temperate regions and Eucalyptus for tropical regions. They are simulated to be harvested every eight years and clear-cut after 40 years. During harvest, 65% of the sapwood and 50% of the heartwood are taken and put into a harvest carbon pool. Herbaceous biomass plantations are represented by the properties of the fast-growing grass types Miscanthus and switchgrass. In contrast to Beringer et al. (2011), these grasses are assumed to allocate carbon on a daily basis and to be harvested as soon as 400 gC/m 2 are reached whereby 85% of the above-ground plant material is taken away (Ashworth, Keyser, Holcomb, & Harper, 2013;Johnson, Clementson, Mathanker, Grift, & Hansen, 2012). Also, bioenergy trees are now parameterized to be more resistant to water stress (through larger rooting depths). Model parameterization results in the best overall match between data and observations. A comparison of the simulated woody and herbaceous plantation productivity with observations from field data verifies that our results capture a realistic magnitude of production . For this study, the LPJmL model is run for the period 1960-2009, forced with the Climate Research Unit's (CRU) TS 3.1 monthly climatology for temperature, cloudiness, and wet days (Harris, Jones, Osborn, & Lister, 2014), and with the Global Precipitation Climatology Centre's (GPCC) precipitation data (Version 5) (Rudolf, Becker, Schneider, Meyer-Christoffer, & Ziese, 2011). If not indicated otherwise, results are presented as 1960-2009 averages. In order to assess the possible bioenergy crop productivity and to inform possible shifts in plantation areas, we assume for our simulations that all bioenergy crops can theoretically be grown everywhere-that is, we derive the theoretical potential for each location and then apply constraints to the actual potential by assumptions about land and water availability (see different scenarios below). The link between bioenergy crop yields and water consumption is simulated for both ends of scale, that is, for a situation without irrigation (rain-fed plantations) and for additional irrigation up to the extent that energy crops do not experience water stress, respectively. Due to the model's ability to separate water fluxes into green and blue the maximum attainable increase of biomass ΔY max related to the amount of blue water consumption ΔWB max -that is, through irrigation-can be calculated by.
where Y tot is the maximum yield achievable if the water demand of energy crops is always fulfilled, Y rf is the yield gained from precipitation only, W GB is the sum of both green and blue water consumption, and W G the green water consumption only (see de Wit (1958); Tanner and Sinclair (1983) and Kiziloglu, Sahin, Kuslu, and Tunc (2009) on linear relationship between water deficit and yield).

| Water withdrawals and availability
We analyze the balance of irrigation water supply and demand at the level of river basins. To calculate water availability (blue water potentially available for irrigation) on river basin level, we aggregate simulated runoff at grid cell level. Thus, we only consider the renewable surface and subsurface water, assuming that no fossil groundwater is available. More precisely, the discharge is first calculated for each grid cell i within a particular catchment area, following the equation.
where D nat i is the "naturalized" discharge (computed based on patterns of anthropogenic land use and land-cover changes (Fader, Rost, Müller, Bondeau, & Gerten, 2010), but without biomass plantations, human water withdrawals, and reservoir storage), D inflow i the incoming discharge from upstream grid cells, R i the local runoff generated in a cell and AET i the actual evapotranspiration from lakes, rivers, reservoirs, and different type of wetlands. In a next step, the "naturalized" discharge was reduced by the amount of water currently abstracted and used to fulfill the demand in the agricultural (W ag ) and nonagricultural (household, industry, and livestock, HIL) sectors, respectively.
HIL is based on recent estimates by Flörke et al. (2013). Finally, we account for environmental flow requirements (EFR) and the limits from seasonal distribution by constraining the maximum availability to 40%-70% of total annual blue water supply (e.g., EFR was set to 30%-60%). Considering different flow regimes and water levels required to maintain the ecological functions (see scenarios described below), we obtain the maximum irrigation water supply of each grid cell, IWS i .
where EFR represents the flow requirement as fraction (0-1) of a grid cell's naturalized discharge D nat i . For this study, we assume that freshwater is not freely distributable within river basins but follows the lateral transport along the river network (Döll & Lehner, 2002). Considering upstream-downstream relationships, the calculated IWS in a cell i does not necessarily correspond to maximum amount of water available for irrigation in that cell. As possible upstream water withdrawals reduce available water downstream, the discharge in any of these connected cells must not become negative. Thus, starting from the most upstream cell, the maximum amount of water that can be abstracted in a cell i corresponds to the minimum IWS of i and all its connected downstream cells, respectively.
Actual irrigation water withdrawal, IWWD, of a grid cell i is determined as.
where A BP i is the area in i potentially available for biomass plantations in the different land constraint scenarios, and EAP is the application efficiency reflecting the conveyance losses of the irrigation system. In case that IWWD i >IWS i , additional water can be taken from the neighboring cell with the largest upstream area, otherwise IWWD i ≤ IWS i . Subsequently, we determine the fraction of the potential plantation area of each grid cell (F BP i ) that could be irrigated to get the maximum yield in that cell.
The additional biomass harvest gained through irrigation (Y IR i ) is calculated as.

| Scenarios of land and water availability
We define four different scenarios for irrigation water supply. The scenario of highest water availability (hereinafter "High") is based on the assumptions that (a) the EFR corresponds to 30% of the annual river discharge; and (b) the irrigation system has no conveyance losses. In the "Medium" water availability scenario, the EFR corresponds to 60% of the annual river discharge while conveyance losses remain zero. In the scenario of lowest irrigation water availability (hereinafter "Low"), EFRs are as in "Medium" but the conveyance efficiency of the irrigation system is T A B L E 1 Scenarios of irrigation water availability

Scenario Definition
High Renewable water resources reduced by current water usage in agricultural and nonagricultural sectors, 30% reserved for ecosystem functions, highly efficient irrigation system (no conveyance losses, withdrawal equivalent to water consumption) Medium Renewable water resources reduced by current water usage in agricultural and nonagricultural sectors, 60% reserved for ecosystem functions, highly efficient irrigation system (no conveyance losses, withdrawal equivalent to water consumption)

Low
As Medium, but less efficient irrigation system with conveyance losses of 25% resulting in higher withdrawal Rain-fed No additional irrigation, precipitation is the only source of plant available water assumed to be 75% which results in a higher withdrawal (gross irrigation requirement). Additionally, we employ a scenario "Rain-fed" where precipitation, that is, green water, is the only source of all plant available water (see Table 1). As shown in numerous publications, future land requirements for food production can be both higher and lower than today (Balmford, Green, & Scharlemann, 2005;Foley et al., 2011;Gerbens-Leenes, 2002;Gerbens-Leenes, Nonhebel, & Ivens, 2002;Hobbs, 2007;Kastner, Rivas, Koch, & Nonhebel, 2012;Ray, Mueller, West, & Foley, 2013;Schmitz et al., 2014;Verburg, Eickhout, & Meijl, 2008;Wallace, 2000). As it is outside the scope of this study to directly model land use and land use change (LUC) associated with future food production, we adopt for our Food first scenario the simple assumption that land currently under agricultural use will not be available for establishment of biomass plantations. For that purpose, spatially explicit global information on cropland extent is obtained from the MIRCA2000 land use dataset (Portmann, Siebert, & Döll, 2010) following Fader et al. (2010).
Further restrictions on land availability are applied to quantify how approaches to protect natural ecosystems and to keep LUC emissions below certain thresholds (based on the estimated carbon storage in soils and aboveground vegetation) influence the plantation supply.
In the Food & Nature scenario, plantations are not allowed on legally protected lands, wetlands, and areas of high biodiversity and/or classified as wilderness. In this scenario, also areas of severely degraded soils are excluded (for detailed description see Beringer et al. (2011)). We acknowledge that this approach does not consider that establishment of biomass plantations can improve conditions for nature and biodiversity. For example, many studies have shown that the integration of new types of biomass production systems into existing agricultural landscapes can have positive impacts by restoring or conserving soils, reducing water pollution, and enhancing landscape diversity (Baum, Bolte, & Weih, 2012;Berndes, Börjesson, Ostwald, & Palm, 2008;Dauber, Jones, & Stout, 2010;Dimitriou et al., 2009;Firbank, 2008;Holland et al., 2015;Manning, Taylor, & Hanley, 2015;Verdade, Piña, & Rosalino, 2015).
Finally, two land use scenarios are defined to represent ambitions to keep LUC emissions below certain threshold levels. The rationale is that low LUC emissions make the so-called carbon payback time (CPT) short. The CPT is the time it takes until bioenergy use has contributed to avoiding as much GHG emissions as was emitted (LUC emissions) when land was converted into biomass plantations associated with the bioenergy system. The CPT depends on the size of LUC emissions and the annual GHG savings arising from the bioenergy use, which in turn depends on the avoided emissions due to displacement of other energy sources and the supply chain emissions of the bioenergy system (Berndes, Ahlgren, Börjesson, & Cowie, 2013;Fargione, Hill, Tilman, Polasky, & Hawthorne, 2008;Gibbs et al., 2008). Thus, larger LUC emissions do not translate into longer CPT if higher annual GHG savings outweigh the effect of the larger LUC emissions. Also, a bioenergy system associated with shorter CTP may not be preferred if the annual GHG savings are small, as the cumulative GHG savings per unit land then grow slowly: preference depends on whether shorter or longer term mitigation effects are priority.
Acknowledging that the level of LUC emissions is an insufficient indicator if the purpose is to evaluate the climate change mitigation value of plantations, we assume that lands are not available for plantations if total carbon stock values are above 150 t C/ha (shrubland/savanna) and 270 t C/ha (forests). Adopting data from Gibbs et al. (2008), these threshold values correspond to a CPT of 10 and 50 years, respectively, if the land is planted with oil palm to produce biodiesel. Table 2 provides a qualitative description of the land use scenarios analyzed here. A quantitative description of the scenarios of water and land availability showing the individual spatial patterns is given in Figures 1 and 2 level, for world regions, at basin level, and at grid cell level, respectively. For reasons of brevity, we focus on herbaceous energy crops. In general, productivity of perennial bioenergy crops species, both C3 trees and C4 grasses, is limited predominantly by water availability (Clifton-Brown & Lewandowski, 2000;Clifton-Brown, Lewandowski, Bangerth, & Jones, 2002). Mainly caused by their higher maintenance respiration, woody bioenergy crops are less productive compared to bioenergy grass (Heaton, Long, Voigt, Jones, & Clifton-Brown, 2004;Ragauskas et al., 2006). However, for comparison, all simulation results for both grass and tree bioenergy crops are presented in Tables S1-S4. Figure 3a demonstrates a large spectrum of plantation supply depending on the chosen combination of land and water availability (Tables 1, 2). The plantation supply in the FF case (2,350 EJ/year) corresponds to a bioenergy plantation area roughly threefold the current global agriculture area (Figure 3c) and a global water use six times higher than currently (Figure 3b). Doubling the ecological flow requirements in this scenario from 30% to 60% of renewable freshwater resources reduces the global supply from irrigated biomass plantations by about a quarter and the total supply (irrigated + rain-fed) by 215 EJ. If 25% of water withdrawn for irrigation of bioenergy plantations is lost during conveyance, the supply from irrigated biomass plantations is reduced by another 13% (total supply, 4%) (Supporting information Tables S1,). The effect of constraining the availability of water resources depends on land availability, for example, a doubling of the ecological flow requirements to 60% of renewable freshwater resources reduces the supply from irrigated biomass plantations significantly more (37%) in the CTS scenario compared to FF. This is because the land constraint applied in the CTS scenario primarily excludes plantation areas with a significant production potential under rain-fed conditions. The further reduction caused by assuming 25% conveyance losses is, however, roughly the same as in the FF scenario (12%). Land availability also influences the relative contribution of rain-fed versus irrigated systems. In the FF and CTS cases, 67% and 16% of the biomass is grown under rain-fed conditions, respectively (water use scenario "High").

| Regional level
The regional contribution to the global plantation supply is shown for each land and water use scenario in Figure 4 (see also for abbreviations of the world regions). Comparing all land use scenarios under rain-fed conditions reflects how these plantation supplies depend on land accessibility and natural environmental conditions. In the FF scenario, the world regions LAM (36%), AFR (21%), and PAS (12%) contribute most to the global plantation supply. An additional exclusion of protected areas (FN) reduces the contribution of LAM to 13%, as large nature reserves such as the Central Amazon Conservation Complex, the Iguazu National Park, and the Cerrado Protected Areas are located in this region. Introducing a carbon threshold as in the CTF scenario reduces the contribution of FSU and PAS by 57% and 67%, respectively, as areas with considerable soil and vegetation carbon stocks (but currently not protected) are widespread in these regions. Constraining land availability further in the CTS scenario once again causes a shift of regional shares. In this most restrictive land availability scenario, more than half of the total plantation supply comes from LAM (24%) and AFR (27%). The contribution of PAO and SAS roughly doubles to 14% and 13%, respectively, while it decreases to <1% for FSU and PAS. The contribution of the MEA region is negligible in all scenarios. Figure 4 shows the regional contribution to the global irrigated plantation supply and reveals distinct regional patterns of dependence on irrigation water availability. The relative contribution of AFR to global plantation supply is consistently larger when irrigation is available. In the other dominating region, LAM, it is smaller in the FF and CTS scenario while it is roughly the same as under rain-fed conditions in the two other scenarios. In AFR in the FF scenario, the relative contribution decreases as irrigation water availability decreases, stays relatively constant in the FN and CTF scenarios, while it increases with decreasing irrigation water availability in the CTS scenario. The same pattern of increasing relative contribution under decreasing water availability is seen for LAM in the FF scenario. One interpretation of this pattern is that-in the absence of land availability restrictions-AFR has relatively large areas where irrigation can boost yield levels substantially, while the irrigated production in LAM is less sensitive to constraints on irrigation water availability. When only unprotected "low-carbon" land is available, it is instead the irrigated plantation supply in AFR that is less sensitive to Forest," (d) "Carbon threshold savanna," as described in Table 2 constraints on irrigation water supply. As a third regional example, results for the CPA region suggest that land availability is limiting the plantation supply in FF, FN, and CTF. However, in the CTS scenario, water availability constrains the plantation supply. In this scenario, land available for biomass plantations is located in basins where the withdrawal limit is either exceeded or available renewable water resources are very low (<10 km 3 ) (compare Figures 1, 2).

| Basin level
The analysis of simulation results at basin level elucidates trade-offs between land and water availability. This is exemplarily illustrated for the FN scenario, showing the plantation supply for the different water availability scenarios ( Figure 5). Biomass supply is in some basins limited by land rather than water availability, that is, it is close to the nonwater limited maximum and not all water resources potentially available for irrigation are used. Examples include basins in large parts of LAM and NAM, EUR, and the southern part of CPA. In contrast, bioenergy supply in most basins in AFR, MEA, SAS, and PAO is constrained by water scarcity. Seven basins were selected to illustrate how the plantation supply changes dependent on availability of irrigation water, and their results are presented in Table 3. In all these basins, simulated maximum plantation supply under irrigation is much larger than the supply under rain-fed conditions. Particularly within the catchment areas of the Syr-Darya and the Colorado, a noticeable plantation supply can only be achieved with substantial irrigation. In the least restrictive scenario concerning water availability for irrigation (High), the Colorado and Sambesi basins could support plantation supplies corresponding to the simulated nonwater limited maximum without exhausting the water resources available for irrigation of biomass plantations. In the five other basins, in contrast, available water resources would become exhausted at plantation supply levels roughly corresponding to 40%-60% of the simulated maximum. In the more restrictive Medium and Low scenarios, available irrigation water resources are exhausted in all considered basins before maximum simulated plantation supplies are reached: for example, plantation supplies in the Colorado and Sambesi basins reach 75% and 55%, respectively, of the theoretical maximum. Plantation supplies in the other basins reach maximally one-third (Medium) and one-fourth (Low) of the water-unlimited potential, respectively.

| Grid cell level
The detailed simulated spatial distribution of biomass plantations can be discerned from Figure 6. Total plantation production within an individual grid cell (Figure 6c) is determined by biomass productivity (Figure 6b) and the share of the grid cell available for biomass plantations (Figure 6a). As shown in Figure 6b, the biomass productivity is relatively high in many regions and the application of irrigation brings about substantial productivity increases over large areas. However, as shown in Figure 6a, low area availability restricts the total plantation supply in several highly productive regions, notably Europe, the Mid-south of the United States, Central America, and the Australian East coast. Large irrigated plantation production occurs in the Mid-south, Southeast, and East coast regions of the United States, in eastern Brazil and northern Argentina, around the Congo basin, and in India. Considerable plantation production is also simulated, for example, in Southern China, Southeast Asia, and the Northwest of Australia (Figure 6c).

| DISCUSSION
We show that availability of land and water, and (irrigation) management of water resources, strongly influence the biomass supply from plantations in individual regions and globally, with water limitations becoming the more relevant, the stricter the assumed constraints on land availability for plantations. This basically confirms conclusions in other studies (Beringer et al., 2011;Creutzig et al., 2015;Offermann et al., 2011) that future possible levels of biomass supply cannot be specified to a narrow range due to inherent uncertainties concerning critical factors, including criteria of exclusion to protect both land and water-which we here distinguish more systematically with respect to their relative effects. Somerville and Youngs (2014) argued that "…estimating the practical limits to how much bioenergy could be produced in the future has been an academic sport in recent years, but is not a useful activity." Slade et al. (2014) reviewed more than 120 estimates of the future contribution of biomass to global energy supply. They noted that the range of estimates is driven more by the choice of alternative assumptions than methodological differences and concluded that studies provide limited insight into the level of deployment that might be achievable in practice, as many open questions will only be resolved as The results presented in this study may in this regard inform the planning of further studies to address aspects of biomass mobilization in basins or multi-basin regions identified as potentially important locations for large-scale biomass production. Such studies can complement scenario studies of the kind summarized by the IPCC (Edenhofer et al., 2011) by providing more comprehensive regionallevel information about current and prospective availability and use of land and water resources, as well as other factors that influence conditions for deployment.
To limit global warming to 2°C (1.5°C), integrated assessment studies project a BECCS contribution of 0.5 to 2.7 GtC/ year (0.6 to 4.1 GtC/ year) from dedicated bioenergy crops to negative emission requirements in 2050 (Fuss et al., 2014;Rogelj et al., 2015). The simulation results reported here do not provide a basis for ruling out bioenergy deployment at levels found in those studies. The higher end in the ranges for bioenergy use in these scenarios is within the range for plantation supply in the most restrictive land use scenario CTS. But the results cannot either be used as a basis for concluding that the scenarios are feasible, as prospects for bioenergy depends on many factors that have not been considered in this study, for example, challenges related to high plantation expansion rates including the build-up of irrigation systems and governance to avoid or mitigate negative effects and maintain public support for bioenergy. According to Turner, Field, Lobell, Sanchez, and Mach (2018), the rate of land use conversion in climate change mitigation scenarios with a likely chance of limiting global warming to 2°C in 2100 proceeds at a median rate of 8.8 Mha/year from 2020 to 2050. In total, the authors estimate 272 Mha of new energy cropland to be enrolled over the next three decades. These results are comparable to our assessments, where in the most restrictive CTS scenario a plantation area of 289 Mha is needed to yield 3 Gt DM/year (that is~1.5 Gt C/year) of highly productive bioenergy crops. If evaluated consequently in terms of environmental impacts such as those considered in the concept of "planetary boundaries," there is little maneuvering space left for biomass plantations Heck, Gerten, Lucht, & Popp, 2018). Rising atmospheric CO 2 content as well as associated climate change effects may affect biomass supply estimates provided here (Deryng et al., 2016;Haberl et al., 2011), yet such impacts need to be examined in the future studies.
Some results of this study, especially regarding the simple assumptions about freshwater availability for irrigation of plantations, are subject to uncertainties. For example, our parameter choice regarding EFRs is based on the hydrologic consideration that water withdrawal in excess of specific percentages of natural mean annual flow represents a risk to riverine ecosystems' integrity. To account for different levels of protection, we specified scenarios representing a range of flow volumes (and irrigation efficiencies) but do not differentiate across regions. In a refined assessment, different EFR estimation methods could be used to assign different EFR volumes depending on month, location, and flow regime. This issue has been addressed in other studies based on a version of LPJmL (developed in T A B L E 3 Utilization of maximum biomass production for selected river basins and under different scenarios of water availability  | 639 parallel to the one employed here) that incorporate dynamic representations of EFRs and also system irrigation efficiencies (Jägermeyr, Pastor, Biemans, & Gerten, 2017), yet solely focusing on agricultural water demand-including bioenergy water demand in an integrated assessment with a more sophisticated modeling approach is a research desiderate for future applications. Meanwhile, a more detailed assessment of uncertainties of supply estimates, with a focus on the choice of freshwater allocation rules within river basins, is subject of a follow-up study (Y. Jans et al., in preparation). The strong influence of irrigation water availability on the simulated level of plantation supply needs to be considered in the context of evolving water scarcities, which are expected to be influenced by both demand-side and supplyside factors where climate change is an important but uncertain factor (Gosling & Arnell, 2016;Haddeland et al., 2014;Mekonnen & Hoekstra, 2016;Schewe et al., 2014). Climate-driven freshwater limitations could in some regions necessitate reversion of large cropland areas from irrigated to rain-fed management, while freshwater abundance in other regions may help ameliorate resulting production losses, if required infrastructure investments take place (Elliott et al., 2014;Jägermeyr et al., 2016Jägermeyr et al., , 2017. The freshwater use associated with the irrigated plantation supplies reported in this study (Figure 3) is very significant in comparison with current freshwater withdrawals (Oki & Kanae, 2006) and consumptive use (Shiklomanov & Rodda, 2004), as well as global freshwater withdrawal limits considered to be associated with physical water scarcity (Defraiture, Molden, Amarasinghe, & Makin, 2001;Vorosmarty, Green, Salisbury, & Lammers, 2000) and planetary boundaries (Rockström et al.., 2009;Steffen et al., 2015). It is, however, not straightforward to translate such observations into firm conclusions about the feasibility of certain levels of biomass supply, as water-related consequences of biomass deployment depend on location, context, and rationales behind the deployment. The cultivation of drought tolerant plants, such as feedstock for bioenergy and other bio-based products, may offer an alternative use of lands where emerging freshwater scarcity causes challenges for irrigated production of more vulnerable crops. Conversely, a possible future need for increased food production in regions with relative water abundancy should be considered if biomass production for bioenergy and other bio-based products is contemplated as an option for making productive use of land in areas where food production is currently not viable due to more competitive production elsewhere. The integration of new forms of biomass production into agricultural landscapes can provide several products from the same land area, enhance resource use efficiency, and help mitigate some of the impacts associated with current land use (Berndes & Fritsche, 2016).