Crassulacean acid metabolism (CAM) offers sustainable bioenergy production and resilience to climate change

Biomass production on low‐grade land is needed to meet future energy demands and minimize resource conflicts. This, however, requires improvements in plant water‐use efficiency (WUE) that are beyond conventional C3 and C4 dedicated bioenergy crops. Here we present the first global‐scale geographic information system (GIS)‐based productivity model of two highly water‐efficient crassulacean acid metabolism (CAM) candidates: Agave tequilana and Opuntia ficus‐indica. Features of these plants that translate to WUE advantages over C3 and C4 bioenergy crops include nocturnal stomatal opening, rapid rectifier‐like root hydraulic conductivity responses to fluctuating soil water potential and the capacity to buffer against periods of drought. Yield simulations for the year 2070 were performed under the four representative concentration pathway (RCPs) scenarios presented in the IPCC's 5th Assessment Report. Simulations on low‐grade land suggest that O. ficus‐indica alone has the capacity to meet ‘extreme’ bioenergy demand scenarios (>600 EJ yr−1) and is highly resilient to climate change (−1%). Agave tequilana is moderately impacted (−11%). These results are significant because bioenergy demand scenarios >600 EJ yr−1 could be met without significantly increasing conflicts with food production and contributing to deforestation. Both CAM candidates outperformed the C4 bioenergy crop, Panicum virgatum L. (switchgrass) in arid zones in the latitudinal range 30°S–30°N.


Introduction
Strategies to improve energy security and mitigate against the impacts of climate change commonly include the production of energy from biomass (EIA, 2013;IEA, 2013). On a calorific basis, the most important biomass resources in future energy scenarios are feedstocks specifically grown for energy (Slade et al., 2014). However, the utilization of land for dedicated energy crops is controversial. Conflicts that exist between land allocated to bioenergy production, food production, and biodiversity conservation have been called the 'bioenergy trilemma' (Tilman et al., 2009). These conflicts are predicted to become more acute with the negative impacts of climate change on water availability and food production, set against the increasing demands of a growing global population (Tilman et al., 2009;Godfray et al., 2010;Van Renssen, 2011;Creutzig et al., 2012;Howells et al., 2013;Wheeler & von Braun, 2013;Slade et al., 2014). One possible solution is to restrict bioenergy production to marginal and low-grade lands. Yet the extent to which such areas may be utilized remains largely unknown and would require improvements in plant water-use efficiency (WUE) that exclude most conventional C3 and C4 bioenergy crops (Somerville et al., 2010;Slade et al., 2014). In this context, we present the first global-scale geospatial productivity model for two bioenergy candidates of the highly water-use efficient crassulacean acid metabolism (CAM) pathway. We also provide an assessment of the resilience of these candidates to climate change and investigate productivity potential on low-grade lands, which we define using Food and Agriculture Organisation (FAO) land-use classifications (FAO et al. 2012).
Crassulacean acid metabolism has evolved on multiple occasions mostly in semi-arid, subtropical habitats and is widely considered to be an adaptation to low and intermittent water availability (Ellenberg, 1981;Ting, 1985;Winter & Smith, 1996). In contrast to the C3 and C4 metabolic pathways, CAM is characterized by the temporal separation of carboxylase activities and a four-phase carbon-uptake pattern over the diel cycle (Osmond, 1978). Phase I (PI) nocturnal atmospheric CO 2 fixation is mediated by phosphoenolpyruvate carboxylase (PEPC) leading to the 4C product malic acid; Phase II (PII) occurs early in the photoperiod and is defined by an overlap in PEPC and ribulose-1, 5-bisphophate carboxylase oxygenase (RuBisCO) activities; Phase III (PIII) occurs behind closed stomata during the photoperiod when CO 2 is supplied to RuBisCO from the decarboxylation of malic acid; and finally, the direct RuBisCO-mediated fixation of atmospheric CO 2 may occur during Phase IV (PIV) towards the end of the photoperiod under wellwatered conditions (Osmond, 1978). The primary uptake of CO 2 at night, when temperatures are low, reduces evaporative demand across the open stomata and is a key WUE adaptation of the CAM pathway (Osmond, 1978;Smith & Nobel, 1986;Nobel, 1988;Winter & Smith, 1996). Water storage parenchyma tissue of succulent CAM species provides a buffer against periods of water deficit stress, and rectifier-like root hydraulic conductivity responses minimize retrograde plantsoil water losses and allow plants to capitalize on short periods of water availability (Barcikowski & Nobel, 1984;Smith & Nobel, 1986;Nobel, 1988).
Succulent CAM tissues allow plants to maintain water homeostasis and facilitate a carbon acquisition strategy of 'drought avoidance' (Borland et al., 2011(Borland et al., , 2014. This is distinct from 'drought tolerance', which is typically observed in arid and semi-arid C3 and C4 plants that show the capacity to endure low cell water potential or in extreme cases, desiccation (Ogburn & Edwards, 2010). Integrated over a 24-h period, typical WUE (defined as the ratio of mmol CO 2 fixed to mol H 2 O lost) values are 0.5-1.5 for C3 plants, 1.0-2.0 for C4 plants, and 4.0-10 for plants displaying CAM (Nobel, 1991). Under ideal conditions, some species of Agave and Opuntia average 43 Mg ha À1 yr À1 above-ground dry mass productivity which is comparable to agronomic C4 species and C3 herbaceous species and trees (Nobel, 1991). These features have lead researchers to propose that CAM plantations could be more resilient to climate change and offer higher productivity on lowgrade and marginal lands than conventional C3 and C4 biomass crops (Borland et al., 2009(Borland et al., , 2014Davis et al., 2011). We tested these hypotheses by (i) constructing a global-scale geospatial productivity model for the CAM biomass candidates Agave tequilana and Opuntia ficusindica, (ii) simulating productivity under present-day and future climate scenarios using outputs from representative concentration pathway (RCP) scenarios presented in the IPCC's 5th Assessment Report (AR5), (iii) applying macro-scale land-use constraints to estimate productivity potential on 'low-grade' lands, and (iv) comparing present-day simulations to outputs of a recently published model for the C4 biomass candidate, Panicum virgatum L. (switchgrass) (Kang et al., 2014).
The model is based upon a refined version of the Nobel environmental productivity index (EPI) methodology, which has been validated at an agronomic scale (Nobel & Valenzuela, 1987;Nobel, 1988). Our refinements facilitate the integration of geospatial data sets to include soil water potential (Ψ s ) as a function of texture class and precipitation, and the capacity of CAM to buffer against periods of low Ψ s . These refinements and the use of multidecadal GIS data sets to estimate, analyse, and predict productivity under various climate change scenarios are a significant advance on the initial use of the EPI approach which used point data from weather stations to predict productivity (Gracia de Cortazar & Nobel, 1990). The two candidates considered in this study, A. tequilana and O. ficus-indica, were selected for their high productivity potential (Nobel, 1991;Borland et al., 2009) and favourable composition for bioenergy conversion processes over the life cycle (Yan et al., 2011).

Refined Nobel environmental productivity index (EPI)
CAM productivity simulations were conducted according to established Nobel EPI methodology (Nobel & Meyer, 1985;Nobel & Quero, 1986;Nobel & Valenzuela, 1987;Nobel, 1988;Garcia-Moya et al., 2011) with further refinements to accommodate for spatial and temporal fluctuations in ecophysiological inputs. These include soil water retention characteristics, CO 2 uptake persistence during drought, and CO 2 uptake response to contrasting day and night temperatures.
EPI methodology states that CAM biomass productivity may be estimated by the product of three dimensionless ecophysiological response indices that quantitatively describe the effect water (I w ), temperature (I t ), and photosynthetically active radiation, PAR, (I p ) availability on net carbon uptake (Nobel & Valenzuela, 1987;Nobel, 1988Nobel, , 1989. Ecophysiological response indices were calculated at a temporal resolution of 1 month and averaged over a 1-year period according to Eqn (1).
The EPI score was then multiplied by a value for maximum above-ground dry biomass productivity (P m ), which could occur under irrigated conditions with optimum planting-density according to Eqn (2). In this study, P m was taken as 44 and 46 Mg (dry) ha À1 yr À1 for A. tequilana (Nobel, 1988;Yan et al., 2011) and O. ficus-indica (Nobel et al., 1992), respectively. Although EPI methodology does not explicitly link plant biochemistry and physiology to productivity, the multiplication of EPI by P m implicitly takes into account agronomic scaling effects such as leaf shading.

Development of ecophysiological response indices to environmental inputs
Ecophysiological response to water, I w . Plant water uptake is a passive process that occurs when soil water potential (Ψ s ) exceeds plant water potential (Ψ p ). Ψ s is defined as the sum of matric, osmotic, pressure, and gravitational component potentials (Campbell, 1988). Matric potential contributions are the most important determinant of Ψ s across varying soil texture classes (Cosby et al., 1984;Saxton et al., 1986;Sperry & Hacke, 2002) and occur as a result of the cohesion between water molecules and the adhesion of water molecules to the soil matrix. Water-adhesive interactions dominate the hydraulic properties of soils with high specific surface area (SSA) and result in lower measures of Ψ s at a given soil water content compared to low SSA soils (Cosby et al., 1984). Fine clay soils of particle diameter less than 0.002 mm have a SSA up to 840 m 2 g À1 , whereas the SSA for gravel soils of diameter 2 mm may be as low as 0.0005 m 2 g À1 (Cerato & Lutenegger, 2002). Under most conditions, Ψ p fluctuates around À0.5 MPa for Agave and Cacti meaning that water uptake can occur, on average, when soil water potential (Ψ s ) > À0.5 MPa (Nobel, 1988).
The precipitation requirement for Ψ s > À0.5 MPa was estimated for each United States Department of Agriculture (USDA) soil texture class identified in the Harmonised World Soil Database (HWSD) World Soil Atlas (see Data S1). Soil water potential was calculated as a function of soil water content (h) and texture class according to Eqn (3) (Saxton et al., 1986).
In Eqn (3), A and B are coefficients that describe soil texture contributions to Ψ s as function of sand (particle size 0.05-2.0 mm) and clay (particle size <0.002 mm) content on a mass/mass basis (Saxton et al., 1986). An approximation of h at Ψ s = À0.5 MPa was found for the 13 texture classes identified in USDA standards by transposing Eqn (3) to give Eqn (4).
Soil water potential was estimated as a function of precipitation and texture class by considering that the precipitation required to elevate Ψ s to À0.5 MPa is proportional to h at Ψ s = À0.5 MPa. Experimental data show that the relationship between days per year when Ψ s > À0.5 MPa at 100-150 mm below the surface (approx. root depth of Agave and Cacti) and precipitation is approximately linear (Nobel, 1988). This allowed a gradient function (g i ,) to be developed to estimate the duration in days when Ψ s > À0.5 MPa as a function of precipitation and soil texture class, i. Values for g i , together with soil texture characteristics, are given in Data S1. Geospatial raster files for g i were then constructed using soil texture class GIS data sets available from the HWSD (FAO et al. 2012) shown in Data S1. The number of days per month (U days ) when plant water uptake could occur (Ψ s > À0.5 MPa) was calculated as the product of g i and precipitation according to Eqn (5) where R is precipitation.
The equations used to estimate soil water potential are valid for a wide range of textures and values of h under unsaturated conditions (Saxton et al., 1986).
The value for U days was then scaled according to the capacities of both A. tequilana and O. ficus-indica to buffer against periods of water stress through incorporating a drought resistance factor (F d ). F d was taken as the fraction of carbon assimilation persistence after the onset of drought (A d , Ψ s < À0.5 MPa) divided by carbon assimilation under optimal conditions (A o , Ψ s > À0.5 MPa) integrated over 28 days according to Eqn (6).
Values for F d were calculated from TA responses to water deficit (Acevedo et al., 1983;Nobel & Valenzuela, 1987) and further validate against eddy covariance gas exchange data (N. Owen, unpublished data). For A. tequilana, F d = 1.37 dmnl and for O. ficus-indica, F d = 1.92 dmnl. These factors indicate that cumulative CO 2 uptake after 1 month of drought was 37% and 92% of uptake under optimum conditions for A. tequilana and O. ficus-indica, respectively.
The effective number of days per month (U e ) when plantcarbon uptake is not rate-limited by water availability was determined by Eqn (7).
The ecophysiological response index for water, I w , was taken as the fraction of effective days where plants could uptake carbon divided by the number of days in the month (D m ) according to Eqn (8). where Ecophysiological response to temperature, I t . The index for carbon-uptake response to temperature, I t , was similarly determined from TA response to minimum (t min ) and maximum (t max ) temperature. The separation of t min and t max inputs was necessary because succulent CAM plants display an asymmetric sensitivity to nocturnal temperature (Nobel & Hartsock, 1978;Medina & Osmond, 1981;Buchanan-Bollig et al., 1984). Minimum-maximum temperature separation also allowed all combinations of day and night temperature to be used as a model input. Low night temperatures facilitate carbon uptake during periods of low evaporative demand, and effect tonoplast permeability, PEPC carboxylation, and malic acid inhibition kinetics (Nobel & Hartsock, 1978;Medina & Osmond, 1981;Buchanan-Bollig et al., 1984;Kliemchen et al., 1993;Carter et al., 1995;Nimmo, 2000). In a similar approach to Owen & Griffiths (2014), carbonuptake responses to incremental changes in day-night temperature data (e.g. 10/25, 15/30, 20/35°C) were weighted in proportion to the fraction of PI nocturnal and PII and PIII diurnal integrated gas exchange under optimal conditions. The combined temperature index, I t , was taken as the average of indexes for t min and t max . Productivity range was restricted to areas where average monthly minimum temperature >0°C, which is consistent with the cold tolerance of both species considered (Nobel & De la Barrera, 2003;Escamilla-Treviño, 2011). Temperature response equations that are given in Eqns (9-12) were derived from TA-response data (Nobel & Valenzuela, 1987;Nobel, 1988;Nobel & Israel, 1994). The fraction of nocturnal uptake (f n ) and photoperiod uptake (f p ) are given in parentheses.
The combined temperature index, I t , was taken as the average of indices for t min and t max according to Eqn (13).
Input data sets and data processing. Ecophysiological response indices were developed according to Eqns 3-15 using CO 2 gas exchange and titratable acidity data presented in Nobel & Hartsock (1983), Nobel & Valenzuela (1987), Nobel (1988), Nobel & Israel (1994), and Acevedo et al. (1983). WorldClim data averaged over the period 1950-2000 was used to estimate current productivity (WorldClim, 2014). Productivity in the year 2070 (P 2070 ) was forecasted using output averaged over the period 2061-2080 from all global climate models (GCMs) models cited in the IPCCs 5th Assessment Report (IPCC et al., 2013;WorldClim, 2014). Geospatial inputs for PAR were estimated from National Aeronautics and Space Administration (NASA) data sets for monthly 'Insolation Incident On A Horizontal Surface' (kWh m À2 day À1 ) averaged over a 22-year period (July 1983-June 2005) (NASA, 2014). PAR (mol m À2 day À1 ) was estimated from solar insolation (I s in kWh m À2 day À1 ) by assuming that 48% of solar energy falls within the PAR range of 400-700 nm (Britton & Dodd, 1976) and a solar radiation to photon flux conversion coefficient of 4.57 mol photons MJ À1 (Amthor, 2010). Data were resampled to 0.1 decimal degrees (DD) using cubic convolution interpolation. World soil texture class data were sourced from the HWSD (FAO et al. 2012).
Sustainable biomass standards set out by the Global Bioenergy Partnership (GBEP), the Roundtable on Sustainable Biomaterials (RSB), and the Renewable Energy Directive 2009/28/EC of the European Union were use as a guide to develop macroscale geospatial constraints to estimate low-impact biomass potential (GBEP, 2011;RSB, 2013). It should be noted that criteria set out by sustainable bioenergy initiatives also commonly include localized factors such as subsistence farming land use, economic viability, and cultural importance (Scarlat & Dallemand, 2011;Dauber et al., 2012). These considerations were beyond the scope of the macro-scale evaluation presented here. The definition of low-grade lands adopted in this study excludes forests, protected areas, wetlands, and highly productive and irrigated agricultural lands. These lands were identified using United Nations Food and Agriculture Organisation (FAO) geospatial land-use data and classifications (Nachtergaele & Petri, 2008). A world map of 'low-grade' lands according to our definition, as well FAO land-use patterns, is given in Data S2. Commercially viable yields were considered to exceed 5 Mg ha À1 yr À1 and assumed to have a dry biomass energy content of 18 GJ Mg À1 (Slade et al., 2014).
Simulated data for switchgrass productivity presented in Kang et al. (2014) were provided by Dr Shujiang Kang. Geospatial data sets were processed using ArcGIS software version 10.1 (ESRI, 2012) at a spatial resolution of 0.1 decimal degrees.

Global biomass yield
World productivity simulations under 20th century conditions for (a) A. tequilana and (b) O. ficus-indica are given in Fig. 1. Productivity ranged from 0.0 to 38.0 (EPI, 0.0-0.86) and 0.0 to 40.0 Mg (dry) ha À1 yr À1 (EPI, 0-0.87) for A. tequilana and O. ficus-indica, respectively. As neither candidate achieved maximum theoretical yield (EPI = 1), these results show that carbon uptake was rate-limited by the seasonal availability of either (or combination of) water, temperature, and PAR in all parts of the World at some time throughout the year. In general, simulations show that O. ficus-indica displays an extended range of high productivity (>50% P m ) compared to A. tequilana. This can be seen across arid and semi-arid regions in Australia, East India, sub-Saharan Africa, and the southern United States. The productive range of both candidates is restricted to the minimum monthly temperature isotherm of 0°C, which is consistent with the cold tolerance of both species.
Although O. ficus-indica generally achieves higher productivity than A. tequilana (max. 40 vs. 38 Mg (dry) ha À1 yr À1 ) and better yield distribution, the higher water-soluble carbohydrate composition of A. tequilana is more suitable for bioethanol conversion processes (Stintzing & Carle, 2005;Li et al., 2012). Therefore, in areas where both candidates produce significant yields, the decision of which to cultivate may in large part be determined by the intended application.
Raster file EPI data of Fig. 1 that are unscaled with P m , are available for download in Data S5 and S6. These maps can be viewed, refined, or rescaled with different values for P m as desired using ArcGIS software (see Data S4 for instructions).

Resilience to climate change
Productivity simulations were performed in the year 2070 under the representative concentration pathway (RCP) scenarios defined in the IPCC's Fifth Assessment Report (AR5). The impacts of climate change were evaluated in terms of change in geospatial and latitudinal productivity distribution, total yield, and the distance that the mean of the latitudinal productivity distribution migrates relative to the equator under 70RCP2.6-8.5 (Figs 2 and 3). The RCPs consider radiative forcing scenarios of 2.6, 4.5, 6.0, and 8.5 W m À2 relative to preindustrial levels by the year 2100 (WorldClim, 2014). The best case scenario is RCP 2.6 (70RCP2.6), and the worst case scenario is RCP 8.5 W m À2 (70RCP8.5).
Simulated change in productivity relative to the present for each RCP scenario is given in Fig. 2. Note that percentage change in yield has been given relative to the present, and error is more significant in areas that currently support very low yields, such as North Africa. Also, while Fig. 2a,b shows that productivity is likely to decline in many areas relative to the present, forecasts relative to the 'best case' 70RCP2.6 suggest a higher capacity for resilience. Climate scenario evaluations against the 'locked-in' best case (between 70RCP2.6 and 70RCP8.5) have been provided in Fig. 2c,d as these simulations may be more applicable to inform policymakers.
Under RCPs 70RCP2.6-6.0, productivity losses for A. tequilana range from À5 to À25% in equatorial regions with isolated areas of resilience and higher productivity restricted to mountain ranges that allow progressive altitudinal migration to locations with lower night temperatures. Outside the latitudinal range of 25°S-25°N, productivity is likely to increase under 70RCP2.6-6.0. The worst case 70RCP8.5 scenario may result in significant losses of >25% in equatorial regions extending to 30°S-30°N. Opuntia ficus-indica shows greater resilience to climate change with losses restricted to À5% to À15% under 70RCP2.6-6.0 across all productive areas and increasing to À25% to À15% in equatorial regions under 70RCP8.5. For A. tequilana, higher night temperatures generally improve yields outside latitudes of 30°S-30°N, but have a negative impact within this latitudinal range. Higher night temperatures extend the productivity range of both species, although suboptimal conditions in these fringe regions only sup-  . Areas of high productivity for A. tequilana are comparatively more restricted than for O. ficus-indica. This is due to the higher sensitivity to minimum temperature and a reduced capacity of A. tequilana to buffer against low soil water potential (F d ). The higher PAR requirement of O. ficus-indica compared to A. tequilana (35 and 29 mol m À2 day À1 , respectively) has a negative impact on yields outside 30°S-30°N. For both species, the productive range is restricted to areas where t min > 0°C. Relative to the best case 70RCP2.6 scenario, however, the outlook is significantly more positive. For A. tequilana, 2070 simulations remain relatively constant under 70RCP4.5-6.0. Opuntia ficus-indica is highly resilient under all climate change scenarios relative to 70RCP2.6, with moderate À5% to À15% losses occurring in some equatorial areas under 70RCP8.5. For both candidates, the greatest losses are associated with higher minimum night temperature and altered precipitation patterns in tropical and subtropical zones. These areas, that are the most vulnerable to the negative impacts of climate change, are mostly excluded from the analysis of lowgrade land yields given below.
Continental-scale climate impacts are considered in Fig. 3a-d. Continents were divided into regions north and south of the equator to investigate yield migration patterns and changes in total productivity under the IPCC climate scenarios. The landmass of Africa north of the equator (Af.N) displayed strong migration resilience for both candidates although productivity is likely to decline by 15-20% and 2-10% for A. tequilana and O. ficus -indica, respectively, for 70RCP2.6-8.5. Areas of Africa south of the equator (Af.S) show considerable resilience with mean yields migrating 30-50 km south for both candidates and decreasing by 10-20% for A. tequilana and 2-8% for O. ficus-indica. Productivity in Asia north of the equator (As.N) is likely to remain constant for A. tequilana and increase by 5-6% for O. ficus-indica, while mean yields migrate by 150-250 and 90-200 km north, respectively, for scenarios 70RCP2.6-8.5. In Asia south (As.S), productivity is likely to decrease by 20% for A. tequilana and 10% for O. ficus-indica with mean yield migration staying relatively constant. Simulations for Australia and Oceania south (A.&O.S) indicate significant productivity resilience although mean yields are likely to migrate 100-160 km for A. tequilana large migration distances are accompanied by significant increases in yield, which is attributed to the extended range where both candidates may be cultivated due to elevated average minimum temperatures. In terms of yield, potential, and migration, the most detrimental effects of climate change are likely to be observed in South America. Latitudinal productivity distribution under present conditions and 70RCP8.5 (Fig. 3c-d also indicate that climate change is likely to have a beneficial impact at latitudes outside 30°S-30°N for A. tequila and outside 25°S-25°N for O. ficus-indica. Mean yields are likely to migrate towards the poles for both candidates. Based on the simulations shown in Fig. 2, O. ficus-indica is consistently more resilient to the impacts of climate change than A. tequilana.

Discussion
In this section, the impact of ecophysiological responses on yield distribution is discussed with the aid of the sensitivity analysis presented in Fig. 4. Sources of uncertainty and key assumptions are then identified and discussed. Productivity simulations for both CAM candidates are then compared to a recently published model of the C4 bioenergy candidate, Panicum virgatum L. (switchgrass). Finally, commercially viable and macro-scale sustainability constraints are applied to out-puts presented in Fig. 1 to estimate viable low-grade land (VL) yield potential. Figure 1 shows that O. ficus-indica is likely to have a more extensive range of high productivity potential (EPI > 0.5) than A. tequilana. This is attributed to the comparatively greater capacity to buffer against periods of low soil water potential (described by factor, F d ) and lower sensitivity to minimum temperature (t min ). For the conditions defined in the sensitivity analysis in Fig. 4, increasing F d by 20% results in an approx. 20% and 10% increase in carbon uptake for A. tequilana and O. ficus-indica, respectively. This shows that an increase in F d would benefit A. tequilana more than O. ficusindica. The greater capacity of O. ficus-indica to buffer against water deficit also meant that a 20% increase in precipitation has a smaller positive impact on yield than for A. tequilana in areas that receive approx. 50 mm of precipitation per month. These results suggest that, although both plants show reasonably strong resilience to climate change, the amount and temporal distribution of precipitation patterns could have a significant impact on yields.

Ecophysiological responses and yield distribution
Soil clay content, which by extension determines soil water potential according to Eqn (4), also has a significant and similar effect on yield as precipitation. A reduction in soil clay content had a more positive impact on simulated productivity for A. tequilana compared to A. ficus-indica due to this plants comparatively lower capacity to buffer against low Ψ s . The impact of clay content on yield, together with soil texture information given in Data S1, indicate that soil properties are a dominant factor determining productivity at a localized scale. In areas where Type 13 clayey soils (75% clay) border Type 1 sandy soils (5% clay), for example, simulated productivity may decrease by up to 80%. As the water-uptake threshold for both species occurs at the same soil water potential (Ψ s = À0.5 MPa (Nobel, 1988)), differences in water relations were determined by F d rather than plant-soil uptake relations.
The sensitivity analysis in Fig. 4 shows that O. ficusindica is significantly more resilient to deviations from the optimum minimum night temperature than A. tequilana. For example, a deviation of 5°C from the optimum reduces simulated yields by 20-33% for A. tequilana compared to a reduction of 2-12% for O. ficus-indica. In general, the negative impacts of climate change in equatorial regions are mostly attributed to high suboptimal t min . At latitudes outside 30°S-30°N, higher t min has a strong positive impact on yield for A. tequilana. However, a recent study on Agave angustifolia showed strong commitment to CAM and high CO 2 (a) (b) Fig. 4 Sensitivity analysis of key input parameters. The sensitivity analysis for (a) Agave tequilana and (b) Opuntia ficusindica shows the effect of changing key input parameters on simulated productivity. The range of the sensitivity analysis for each parameter is shown in the parentheses (lower, middle, upper) as follows: photosynthetically active radiation PAR (20, 25, 30 mol m À2 day À1 ); precipitation (40, 50, 60 mm month À1 ); average minimum temperature, uptake persistence over a wide range of day and night temperatures (Holtum & Winter, 2014). The results of this study suggest that Agave species that inhabit tropical regions, with characteristically higher nocturnal temperatures, could be more suitable for biomass production in warmer equatorial climates. Data used to construct the ecophysiological response index to PAR show that carbon-uptake response saturates at 35 mol m À2 day À1 for O. ficus-indica, higher than for A. tequilana at 29 mol m À2 day À1 (Nobel & Hartsock, 1983;Nobel & Valenzuela, 1987). This has a small negative impact on O. ficus-indica productivity outside latitudes of 30°S-30°N.

Uncertainty and key assumptions
Four categories of potential error were identified: input data sets, methodology, key assumptions, and exclusions. Input data sets include global climate models (GCMs), the harmonized world soil database (HWSD) texture data, solar radiation data, and land-use constraints. To minimize error from GCM inputs, the mean output of all GCMs referenced in the IPCC's 5th Assessment Report averaged over the period 2061-2080 was used to generate forecasts for the year 2070. GIS soil data sets were resolved into 13 categories according to USDA standards to minimize error from soil texture inputs. It should be noted that simulations are subject to the sparsity of data and interpolation methods used to construct all geospatial input data sets (see referenced data for information on input data set limitations).
Methodological error may include error associated with the EPI approach, the use of titratable acidity (TA) to derive ecophysiological response indices, and the scaling of EPI with a value of maximum dry biomass productivity (P m ). Although the Nobel EPI approach has been validated against biomass accumulation and leaf-unfurling rates in the field (Nobel & Valenzuela, 1987;Nobel, 1988Nobel, , 1989, further validation across contrasting environmental conditions would be desirable. The use of TA as proxy to estimate carbon uptake is based on the 1 : 2 : 1 stoichiometric relationship between nocturnal CO 2 fixation, H + and 4C product, malic acid (Osmond, 1978). It should be noted, however, that DTA-based approaches do not account for PIV carbon uptake and PI refixation of respiratory CO 2 and could therefore be prone to error. In the case of A. tequilana and O. ficus-indica, PIV only accounts for approximately 13% and 2% of integrated daily carbon uptake, respectively, under optimal conditions (derived from data presented in Nobel & Valenzuela (1987) and Nobel & Israel (1994)). Additionally, a recent agronomic-scale eddy covariance (EC) gas exchange study over a field of A. tequilana (N. Owen, unpublished data) showed that whole-plant averaged DTA measurements closely agreed with integrated daily gas exchange after 70 days of drought. The EC data support the use of DTA as a proxy for carbon uptake as contributions from recycled respiratory CO 2 refixation in PI (that contributes to DTA) seem to approximately equate to Phase IV CO 2 uptake (which is not measured in DTA measurements) overall under field conditions. The values adopted for P m used to scale EPI to estimate actual productivity may also contribute error. Under the relatively low planting densities employed by current commercial applications (food and alcohol production), P m is approx. 25 and 18-20 Mg (dry) ha À1 yr À1 for Agave and Opuntia species, respectively. However, under optimal planting densities reported, P m is significantly higher, at 38 and 47 Mg (dry) ha À1 yr À1 , respectively, (Nobel, 1991). As P m is dependent on cultivation practices employed, unscaled EPI maps have been provided in Data S5 and S6 which may be rescaled with a chosen value of P m as desired.
Two key assumptions were required to integrate geospatial data sets with plant ecophysiological response data. Firstly, the continuum of night-day temperatures was incorporated by making the assumption that minimum temperature sensitivity is proportional to nocturnal carbon uptake and that maximum temperature sensitivity is proportional to photoperiod uptake of CO 2 . This assumption was based on the well established hypersensitivity of succulent CAM species to night temperature for the species considered (Nobel, 1976;Nobel & Hartsock, 1978;Medina & Osmond, 1981;Acevedo et al., 1983;Grams et al., 1997;Borland et al., 1999). However, a recent study on Agave antustifolia that showed high resilience to temperature suggests that this approach is most likely conservative (Holtum & Winter, 2014). Secondly, the duration for which Ψ s was greater than Ψ p was benchmarked against measured data sets showing linear relationships between precipitation and the duration for which Ψ s > À0.5 MPa (Nobel, 1988). Exclusions from the model include soil-nutrient inputs, species invasiveness, fertilization effects of high atmospheric CO 2 concentration, and plant-acclimation capacity. The latter two are likely to improve CAM yields (Nobel & Israel, 1994).

Comparison between CAM candidates and the C4 bioenergy candidate, Panicum virgatum L (switchgrass)
A geospatial comparison between simulated output for A. tequilana and O. ficus-indica with the outputs of a recently published model of Panicum virgatum L. (switchgrass) productivity (Kang et al., 2014) is given in Fig. 5. Panicum virgatum L. is a perennial C4 bioenergy candidate that has demonstrated high productivity potential over contrasting nutrient, temperature, and water availability regimes (Sanderson et al., 1996;Keshwani & Cheng, 2009). In Fig. 5, only 'commercially viable' yields of ≥5 Mg (dry) ha À1 yr À1 are considered. In general, simulations show that P. virgatum L. achieves higher productivity than both CAM candidates in regions outside the latitudinal range of 30°N-30°S. Inside this latitudinal range, however, A. tequilana outperformed P. virgatum L. with the exception of parts of equatorial South America, Africa where A. tequilana yields are negatively impacted by high average monthly minimum temperature, t min (see Fig. 4). O. ficus-indica generally outperformed P. virgatum L. in almost all areas inside 30°N-30°S with the exception of restricted parts of the wet tropics and northern China where simulated yields were similar (<25% difference).
The different yield-distribution patterns reflect the different carbon acquisition strategies of arid and semiarid perennial grasses compared to succulent CAM species. Herbaceous perennials such as P. virgatum L. have a short growth season that is restricted to warm and wet months from a rootstock that becomes dormant over a dry season or winter period (Skinner & Adler, 2010). These traits allow significant levels of productivity to extend to areas that experience harsh winter conditions at latitudes up to 55°N. On the other hand, O. ficus-indica and A. tequilana grow throughout the year and are highly sensitive to t min . The higher productivity potential of both CAM species in arid and semi-arid areas inside the latitudinal range of 30°N-30°S is mostly attributed to rectifier-like root hydraulic conductivity responses that allow plants to capitalize on low and infrequent precipitation events. These hydraulic responses also allow the recharge of water storage parenchyma which supports continued physiological and metabolic function during extended periods of drought (Ψ s < À0.5 MPa). The results presented here suggest that the different strategies employed by succulent CAM species, compared to C4 grasses, are likely to translate to higher productivity potential in tropical areas of low rainfall.

Viable low-grade land yield potential
Exploiting the bioenergy potential of the CAM pathway in a sustainable manner is contingent upon feedstock cultivation on low-grade land. However, there are no standard definitions for classes of 'surplus' land, which are a function of subsistence farming land use, economic viability, cultural importance, and biodiversity value (Dauber et al., 2012). Such lands must be evaluated at a local scale (Dauber et al., 2012;Immerzeel et al., 2014) and to our knowledge no global-scale geo-spatial data sets that identify categories of surplus land are available. We therefore elected to define 'viable lowgrade' lands (VL) using present-day Food and Agriculture Organisation (FAO) (Nachtergaele & Petri, 2008) land-use classifications. As initial land use is a major determinant of the environmental performance of dedicated bioenergy crops (Immerzeel et al., 2014), areas of high and irrigated agricultural activity, forests, and protected areas were excluded (refer to Data S2 for a list of FAO constraints used). To enable direct comparison with a review of global biomass energy resources, viable yields were considered to exceed 5 Mg ha À1 yr À1 and assumed to have a dry biomass energy content of 18 GJ Mg À1 (Slade et al., 2014). Therefore, VL maps in Fig. 6 illustrate yields that are theoretically possible on 'lowgrade lands' at a global scale, without accounting for local social acceptability or environmental impacts. ASCII files for Fig. 6 have been provided in the Data S7 and S8 to enable adjustment or refinement of the landuse constraints as desired using GIS software. Simulations show that both CAM candidates generally outperform P. virgatum over the latitudinal range 30°S-30°N. Higher seasonal variation in temperature outside this range and in mountainous areas tend to favour P. virgatum L. In the wet tropics, P. virgatum L. and A. tequilana display similar yield potential (<25% difference). Only 'commercially viable' yields ≥5 Mg (dry) ha À1 yr À1 are shown.
Simulations presented in Fig. 6 suggest that VL yields are highly resilient to climate change and may have the capacity to meet 'extreme' (Slade et al., 2014) future bioenergy demands of >600 EJ yr À1 . Once again comparing the 70RCP8.5 to the present, climate change may have a positive impact on VL yields in Asia (+3.6%), Europe (+165%), and North America (+18%) for A. tequilana. Similarly, for O. ficus-indica, climate change may have a positive impact in Asia (+8.8%), Europe (+132%) North America (+28%), and South America (+0.1%). See Data S3 for simulation data presented in the barcharts in Fig. 6.
Compared to the present, 70RCP8.5 world VL yields may decrease by À11% (497-444 EJ) for A. tequilana but remained fairly constant for O. ficus-indica at À1% (756-750 EJ). The land area that supports these yields, shown in Fig. 6, is 1950 Mha (av. 14.2 Mg ha À1 yr À1 ) for A. tequilana and 2300 Mha (av. 18.3 Mg ha À1 yr À1 ) for O. ficus-indica. Similarly under 70RCP8.5, VL yields for A. tequilana and O. ficus-indica may be produced on 2070 Mha (av. 11.9 Mg ha À1 yr À1 ) and 2510 Mha (av. 16.6 Mg ha À1 yr À1 ), respectively. For context, this represents 13.9% and 16.8% of the world's terrestrial land area, respectively. These results suggest that climate change will have a small overall negative impact on world VL yield intensity.
A bioenergy review by Slade et al. (2014) found that dedicated bioenergy crops may contribute 22-1272 EJ by 2050 (Slade et al., 2014). The authors considered scenarios >600 EJ yr À1 as 'extreme' and based on assumptions that increases in food-crop yields will significantly outpace demand, 1000 Mha of high-grade agricultural land will be available for bioenergy, population growth will be low, a primarily vegetarian diet will be adopted, and extensive deforestation would be allowed to continue (Slade et al., 2014). For context, current world bioenergy demand is approximately 50 EJ yr À1 and represents 10% of total world energy demand (IEA, 2013). Significantly, the theoretical results presented here show that VL land yields for O. ficus-indica could meet these extreme bioenergy demand scenarios (>600 EJ yr À1 ) without requiring (a) (b) Fig. 6 Viable low-grade land (VL) productivity potential under IPCC RCP climate change scenarios. The geospatial productivity simulations show VL yields under current climate conditions  and the barcharts give total VL productivity under each RCP scenario in the year 2070 for each continent for (a) A. tequilana and (b) O. ficus-indica. VL yields for were constrained to ≥5 Mg ha À1 yr À1 and exclude highly productive and irrigated agricultural lands, forests, and protected areas. Simulations for A. tequilana show that climate change will have a positive impact in Asia, Europe, and North America. For O. ficus-indica, climate change may have a positive impact in Asia, Europe, North America, and South America. Large expanses of Saharan Africa, the Middle East, and Australia were excluded due to the commercially viable yield requirement of ≥5 Mg ha À1 yr À1 . additional high-grade agricultural land and will continue to do so under the worst case climate change scenario. However, while these results underscore the high productivity potential of O. ficus-indica, we do not suggest such large-scale development would be environmentally or socially desirable, or practically feasible.
The model simulations presented here indicate that CAM bioenergy candidates are likely to outperform C3 and C4 bioenergy crops in terms of low-grade land productivity and meeting sustainable bioenergy objectives. Features including nocturnal carbon uptake, rectifierlike root hydraulic conductivity responses to fluctuating Ψ s , and the capacity to buffer against periods of water deficit stress allow high productivity rates on low-grade lands. The yield simulations highlighted the capacity of the CAM pathway to meet future bioenergy demands with minimum resource conflicts. The same WUE features of the CAM pathway that distinguish it from C3 and C4 bioenergy candidates also offer resilience to climate change. Low-grade land simulations comparing the worst case climate scenario (70RCP8.5) with present suggest that total yield will fall by 11% for A. tequilana and remain approximately unchanged for O. ficus-indica in 2070. However, CAM biofuels are not exempt from criticisms made of first generation C3 and C4 bioenergy candidates. While the macro-scale analysis presented here provides a good approximation of CAM bioenergy potential on a global scale, the sustainable introduction of CAM crops will need to be evaluated on a case-bycase basis and in accordance with sustainable bioenergy standards.