New feed sources key to ambitious climate targets

Net carbon sinks capable of avoiding dangerous perturbation of the climate system and preventing ocean acidification have been identified, but they are likely to be limited by resource constraints (Nature 463:747–756, 2010). Land scarcity already creates tension between food security and bioenergy production, and this competition is likely to intensify as populations and the effects of climate change expand. Despite research into microalgae as a next-generation energy source, the land-sparing consequences of alternative sources of livestock feed have been overlooked. Here we use the FeliX model to quantify emissions pathways when microalgae is used as a feedstock to free up to 2 billion hectares of land currently used for pasture and feed crops. Forest plantations established on these areas can conceivably meet 50 % of global primary energy demand, resulting in emissions mitigation from the energy and LULUC sectors of up to 544 \documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\pm$$\end{document}± 107 PgC by 2100. Further emissions reductions from carbon capture and sequestration (CCS) technology can reduce global atmospheric carbon concentrations close to preindustrial levels by the end of the present century. Though previously thought unattainable, carbon sinks and climate change mitigation of this magnitude are well within the bounds of technological feasibility. Electronic supplementary material The online version of this article (doi:10.1186/s13021-015-0040-7) contains supplementary material, which is available to authorized users.


List of Figures
Historical data from the IEA is included for coal, oil, and gas production, and GEA scenario projections [1] for 2100.

Systems Dynamics Modeling
The system dynamics approach to modeling global impact modeling was originally developed by Jay Forrester at MIT in the 1950s [6,7]. An alternative to reductionism, which dissects complex phenomena into component pieces, system dynamics attempts to comprehend entire systems to understand their past behavior and future evolution [8]. The foundational notion of this approach is that structure determines performance, i.e. that the structure of a system is the primary cause for its behaviorproblematic or otherwise [9]. In practice, this means that systems dynamics models do not optimize objective functions to identify dynamically efficient pathways. By integrating historical data sets, FeliX generates insights into the effects of system constructions and resource allotments, but does not spontaneously or endogenously contemplate alternative systems or allotments. In this context, "system" is defined as a collection of interrelated and interacting nodes, or elements, and "dynamics" refers to the temporal evolution of the system as governed by the same interactions. To maintain correspondence with reality, system dynamics models capture as many of the interactions among the elements within their scope as possible. In this way, they provide insight into feedback loops, or the co-dependent evolution of ostensibly separate sectors of dynamic systems. A change in one variable affects other variables in predictable ways over time. However, this effect can subsequently propagate to alter the course of the original variable, and so on. As these changes accumulate, insight can be gained into trends at the variable, sectoral, and global level. In general, these effects are linear, and complex models can be constructed out of a network of elementary interactions. However, special dynamic notions are also incorporated by delays and other nonlinear relations among the system elements. A thorough description of system dynamics components and technique is presented by [7,8,9,10].
The system dynamics approach to characterizing and quantify the functioning of the climate system is well established [11,12,13,14,15,16]. In cases where such relations have not been quantitatively established, group model building sessions have been convened [17,18,19,20]. These foundational efforts provide both the philosophical and functional basis for the FeliX model, as described in this document.

The FeliX Model
FeliX models the effects of new policies and technologies in the context of fundamental, complex interconnections among social, economic, and environmental Earth subsystems in the Anthropocene Era [21,22]. The model consists of differential equations which link stocks-representing resources-and variables to characterize the present state and future development of natural and economic systems, including: population, GDP, land use, energy, carbon cycling, and climate. Feedback loops define the connections among and drive the co-dependent evolution of these systems. All FeliX model historical data, parameters, and results are calculated and reported as global averages.

Purpose
The model combines historical data with results from similar models, impact figures from published articles and sector reports, and expert interviews to calibrate the model to match available historical data between 1900 and 2010 [3,23]. Future developments in the energy sectors are loosely calibrated to the Global Energy Assessment (GEA) [1]. Therefore, for the purposes of this analysis, primary energy profiles are not FeliX model results, but are rather definitional of scenarios. This document describes the major features and linkages of the FeliX model with a focus on the nominal scenario, Business as Usual (BAU ). Page numbers, where listed, refer to the model report and technical documentation [24], which contains a complete discussion of the representation and calibration of each sector, and which is available for download from the model website [25].

Business as Usual (BAU)
BAU seeks to project the development of agricultural, energy, and carbon systems in the absence of perturbation from new policies. Population growth, per capita food and energy consumption, agricultural yields, and land use evolve endogenously through 2100. In this way, BAU serves as a baseline against which the impacts of novel technologies and policies can be measured.
To understand sources of systematic error in BAU assumptions, we define an envelope of plausibility for eleven major model parameters around their nominal values. Parameters governing per capita demand for and supply of food, feed, and fuel are included as sources of error, and the shifts assigned to each parameter are defined in the relevant section of this report. Treating each parameter as uncorrelated, we run these through the model as independent scenarios. At the end of this report, we use these scenarios to derive estimates of the leading systematic errors of carbon emissions and temperature change projections in BAU. Finally, we use this method to examine leading errors on the emissions mitigation of an accelerated transition to renewable energies (BioEnergy scenario).

Population and GDP
Using a systems dynamics approach, population change is the net effect of fertility and mortality rates. The fertility rate is inversely correlated with educational attainment (p. 252) and GDP growth, and the mortality rate is derived from life expectancy, which is correlated directly with GDP growth and food availability and indirectly with environmental and climactic degradation.
Fertility, life expectancy, and population sub-groups are calibrated simultaneously to FAO data ( Figure S1 A-C) (pp. [26][27]. The World Population Projection Medium Variant and the Shared Socio-economic Pathways (SSPs) are used to validate the resulting baseline population projection (Figure S1 C) [26,27].
The FeliX model uses a neoclassical growth model to characterize the economic sector (p. 11). World gross productivity (GDP) is correlated with labor force, capital accumulation, and the pace of technological developments and calibrated to historical data ( Figure S1D) [28].

Error Analysis
• World Population: The full width of the 80% confidence interval from the World Population Projection is used to define low and high population scenarios, as shown in Figure S1C. • Global GDP: The baseline GDP growth projection is accelerated or retarded linearly between 2010 and 2100 to generate a total cumulative ±20% shift in GDP per capita (relative to nominal) by 2100, as shown in Figure S1D.

Food Demand
Per capita food demand is modeled separately for animal and vegetal calories and calibrated to historical data from the FAO ( Figure S2 at top left). Future demand is scaled between lower limits (10 and 750 kCal day −1 for animal and vegetal, respectively) and upper limits (850 and 3,000 kCal day −1 ) as a function of GDP per capita (pp. 262-263).

Error Analysis
• Food Demand (Animal): Nominal per capita animal food demand is scaled linearly between 2010 and 2100 to generate a cumulative ±10% shift by 2100, as shown in Figure S2 at top left. • Food Demand (Vegetal): Nominal per capita vegetal food demand is scaled linearly between 2010 and 2100 to generate a cumulative ±10% shift by 2100, as shown in Figure S2 at top left.  Table S1 Dimensionless coefficients used to model the competing effects of agricultural yield intensification (INT and land management) and yield-limiting factors. The product of six coefficients, listed in the rightmost column, indicates the scaling factor used to calculate global average cropland yield (see Figure S2 at middle left).
• Feed Percentage from Algae: A range of feed demand reduction due to algae production [10%, 20%, 30%, 40%] is considered in the main paper. For the purposes of this error analysis, 40% of feed demand is nominally met by algae in the Alg-Feed (40%) scenario, and a 50% shift down (to 20% of feed demand) is used in Table S4.

Agricultural Yields
The land use and land use change (LULUC) sector of the model accounts for agricultural intensification due to fertilization, irrigation, and input-neutral technological (INT) advancement, and for de-intensification due to water deficits, climate change, and pollution. Yield scale factors associated with each of these factors are modeled independently (pp. 66-68) and tabulated below.

Error Analysis
• Agricultural Yields: The pace of input-neutral yield growth-nominally a linear extension of historical trends (p. 265)-is shifted up or down to create a ±25% effect in areal agricultural yields, as shown in Figure S2 at middle left. This range is constructed to encompass independent econometric projections of input-neutral yield growth through 2100 [29], indicated by grey bars in the same figure.

Land Use
Agricultural production systems can either intensify of expand in response to growth in demand for feed, food (animal and vegetal) and fuel. In addition to land (pp. 58-74), water (pp. 93-99) input is explicitly modeled. Four types of land are distinguished in the FeliX model: agricultural, forest, urban/industrial, and other (e.g. woodland and grassland). Agricultural land is further partitioned into arable land, permanent crops, and permanent meadows and pastures. Competition among these mutually-exclusive land categories is mediated by demand for the commodities they produce, as discussed above, and by policy restrictions. In particular, demand for food, feed, fuel, and fiber drive the expansion of agricultural land and managed forests at the expense of natural forests and other habitats.
The allocation of agricultural land is shown in Figure S2   and other natural habitats (27% decrease). This expansion is required to meet growing demand for food, feed, and energy as well as to offset anticipated decreases in the productivity of agricultural land due to pollution, exhaustion, and water scarcity.

Error Analysis
• Plantation Productivity The nominal productivity of managed forests (10 dry tons biomass ha −1 y −1 ) is shifted ±50% to cover a range of 5-15 dry tons biomass ha −1 y −1 .
• Energy Crop Productivity The nominal productivity of energy crop land (20 dry tons biomass ha −1 y −1 ) is shifted ±50% to cover a range of 10-30 dry tons biomass ha −1 y −1 .

Energy
Energy demand per capita is calculated as a function of GDP per capita (pp. 45-55) and plotted at left in Figure S3. Supply is modeled independently for each primary source of energy: coal, gas, oil, solar, wind, and biomass ( Figure S4A). Nuclear power is limited exogenously in BAU to present production. The market share of each source is presented in Figure S4B with IEA historical data. Middle-of-the-road GEA projections (GEA med 450 & geama 450 btr full ) for 2100 are indicated in brackets in Figure S4A [1]. Primary energy projections through 2100 are compatible with these moderate GEA pathways circa 2100 with respect to total primary energy production (GEA: 850-909 EJ y −1 vs. BAU : 875 EJ y −1 ), total production minus nuclear and hydroelectric power (GEA: 679-830 EJ y −1 vs. BAU : 779 EJ y −1 ), and fossil fuel usage (GEA: 122-539 EJ y −1 vs. BAU : 482 EJ y −1 ). We also note that FeliX BAU does not include carbon capture and sequestration, while the GEA pathways do.
On top of this calibration, a simulation of price-based competition between energy sectors explicitly models sectoral and global growth and efficiency due to exploration, production, infrastructural investment, R&D activities, and costs of energy carriers. This module is used for impact analysis of exogenous market deformities including renewable resources subsidies and other policy tools.   Figure S3 At left: Energy demand per capita in BAU is a function of GPD per capita. Historical data is derived from energy and population data from the IEA and FAO, respectively. To assign a systematic error to this measurement, the nominal projection is shifted ±0.25% annually after 2010. This leads in 2100 to -20% lower and +25% higher energy demand per capita, as indicated by the red shaded region. At right: Total primary energy demand and production.
As shown in Figs. S3 and S4, renewable sources of energy including solar, wind, and biomass are projected even in BAU to undergo significant expansion. As a result, the market share of fossil fuels is projected to fall below 60% of primary energy supply by 2100, even as absolute consumption rises 17% relative to 2010. Over the same period, biomass production grows to satisfy nearly a quarter of global energy demand.
In the BioEnergy scenario (cf. Fig. S4), absolute fossil fuel consumption falls 12% in 2100 relative to 2010. Over this period, the market share of renewable energies expands to 60% of total primary energy production.

Error Analysis
• Energy Demand Per capita energy demand is shifted by a factor of ±0.25% y −1 from 2010-2100 to simulate on the low side gains in energy use efficiency  and, on the high side, increased demand. This geometric approach shifts total energy demand asymmetrically, with low demand at 701 EJ y −1 (-20% relative to nominal) and high demand at 1091 EJ y −1 (+25%) in 2100.

Carbon Cycling
The climate sector of the FeliX model is based on the C-ROADS model [30], which in turn refers to the FREE [16,31] and DICE [32,33]

Error Analysis
• Biomass Fixed Emissions Nominal net emissions from bioenergy (0.05 tC tDM −1 ), which reflect transportation and processing energy costs, are shifted ±100%. • Agricultural Emissions Nominal agricultural emissions (including those from fertilizer)-which total 1.12 PgC in 2010 and 2.16 PgC in 2100-are shifted ±20%. • Forest C Sequestration The carbon sink of standing forests (nominally 85 tC ha −1 ) is shifted ±25%. This affects the carbon penalty for deforestation as well as the incentive for afforestation.

Climate
The climate sector of FeliX draws out the effects on climate of the accumulation of carbon dioxide (CO 2 ), methane (CH 4 ), nitrous oxide (N 2 O), hydrofluorocarbons (HFCs), and other greenhouse gases. Specifically, it simulates the warming on the surface of the Earth and in the upper ocean due to these emissions in accordance with [31] and [33]. The Representative Concentration Pathway projection of 4.5 W/m 2 is used for all radiative forcings except CO 2 , which is determined endogenously. Positive forcing increases the atmospheric and upper ocean temperatures, as shown in Figure S7, and the transfer of carbon and heat into deeper ocean layers is also modeled explicitly (pp. 84-91). The consequences of climate change are subsequently propagated through all sectors, with consequences for land fertility, population growth, and biodiversity, among other parameters (p. 89).

Error Analysis
• Non-CO 2 Emissions: RCPs for CH 4 , N 2 O, HCFs, and "other" greenhouse gases are shifted down (RCP 2.6) or up (RCP 8.5) simultaneously to model the warming effects of low and high non-CO 2 emissions pathways, respectively.

Algae Costs
The operating costs of the Algae Energy Farm in Queensland Australia are listed in Table S2. Capital expenditures are assumed to be amortized over 20 years at 3% interest y −1 . Carbon dioxide is the single most expensive input, but co-location with industrial carbon streams and the expansion of CCS infrastructures would make it feasible for algae farms to be paid to consume these emissions.

Agricultural Yields and Land Use
Projected yields and avoided agricultural land use change in the Alg-Feed (40%) scenario (relative to BAU ) are plotted in Fig. S8. Areal yields are higher in the Alg-Feed (40%) scenario due to reduced competition for water (irrigation), but this effect is smaller than the error envelope assumed for agricultural yields. Time series

Emissions
Projected annual emissions for five major fuel sources and land use & land use change (LULUC) in the BAU, BioEnergy, Alg-Fuel and Alg-Feed (40%) scenarios in 2100 are shown in Table S3. In the Alg-Fuel and Alg-Feed (40%) scenarios, total emissions are also tabulated after CCS is used to capture 25%, 50%, or 75% of annual emissions from the energy sector. Atmospheric carbon concentrations are plotted in S10.
Figure S11 displays (left) projected net annual emissions and (right) temperature anomalies in year 2100 of the FeliX simulation for a range of CCS efficiencies (gross energy sector emissions reductions) and levels of algae production (as percentages of total feed demand).

Sensitivity Analysis
An envelope of plausibility has been defined for eleven FeliX model parameters. In all cases, the range of this envelope is reasonably conservative, as defined relative to the RCPs, SSPs, and other similar projections. Most importantly, this range establishes the relative magnitude of the effect that each parameter has on emissions projections. The results of this sensitivity analysis on emissions savings in Alg-Feed (40%) relative to BAU are shown in Table 2 of the main paper. Absolute effects on the BAU and Alg-Feed (40%) scenarios are listed in Tab. S4. For each parameter, two additional scenarios are defined in which the parameter is shifted above or below its nominal value (as described in Error Analysis subsections throughout). The impacts of each shift on total cumulative emissions [2010-2100] is calculated and shown relative to nominal values (at top) in Table S4.
Alternative per capita energy demand and population pathways have the largest effects on cumulative emissions (±8% and ±6%, respectively) in the BAU scenario. Emissions reductions in the Alg-Feed (40%) scenario are dependent on several additional factors or assumptions, including especially biomass fixed emissions and plantation productivity. Figure S10 Time series of atmospheric CO 2 concentrations. Dark shaded ranges show the effects of population growth on the BAU and Alg-Feed (40%)+CCS projections, and the lighter range depicts sensitivity of the latter scenario to energy crop land productivity. For comparison, the four IPCC RCPs are also displayed [4]. Historical data from CDIAC used for validation [5].