Global biomass feedstock supply
We begin with inter-model perspectives on the global supply of biomass, specifically evaluating the distribution of supply. Figure 1 shows the least-cost regional and feedstock supply distributions from each of the models for each of the four global biomass supply scenarios. The color indicates the region of supply, while the color shading and hatching indicate the biomass feedstock type. We immediately observe significant variation between models in the distribution of biomass feedstock production, with the distributions changing over time in all models, but the patterns very different. At lower levels of global feedstock demand (“B100,” 100 EJ/year by 2100), we find residues representing a larger fraction of the supply, while at higher levels of demand (e.g., “B400,” 400 EJ/year by 2100), energy crops and managed forest are more prominent. Thus, when modeled, residues are considered a lower cost feedstock that would dispatch first, but are of more limited supply than energy crop and managed forest feedstocks. All models exhibit greater regional diversification of feedstock production as global supply increases (over time and across scenarios); but, in some models, we observe strong regional preference for supplying biomass. For instance, in DNE21 +, Asia is a more prominent source of biomass than in the other models, while the key supplying regions are the Middle East and Africa for the AIM model, Reforming Economies for FARM, the OECD for GCAM, and Latin America for IMAGE. As for feedstock type, energy crops are a significant share of supply in all the models, while residue supplies are most prominent in BET, GRAPE, and IMAGE, and managed forest biomass supplies are cost competitive in the three models modeling forest feedstocks as an option—BET, GLOBIOM, and NLU.
All the models, except one (NLU), exhibit increasing net land CO2 emissions globally as the demand for modern biomass increases (Fig. 2). However, for a given supply scenario, there are significant differences in the cumulative CO2 emitted, with BET, IMAGE, and MAgPIE estimating the greatest net land-related CO2 emissions, while DNE21 + , FARM, and NLU estimate the least. The results are indicative of the types of biomass feedstocks modeled, the relative cost-effectiveness of different feedstocks, and the characteristics of the locations supplying the biomass in terms of land productivity, land conversions, and carbon density assumptions. For instance, IMAGE and MAgPIE are primarily supplying energy crops, but converting larger amounts of forest and other natural lands than the other models (see discussion later on the differences in the average characteristics of supply between models). The models also vary in the sensitivity of the CO2 emissions to the level of biomass supply, with some models relatively insensitive (e.g., NLU, FARM), while all the other models project more than a doubling of CO2 emissions from a doubling of biomass supply. These results are representative of the shifts in feedstock type and source location as supply increases, as well as simply the increase in the global amount of biomass being supplied. Consistent with the differences in the distributions of feedstock supply (Fig. 1), we see large variation in regional CO2 emissions for a given biomass supply level, as well as the sensitivity across levels.
For some models, regions, and supply levels, reductions in land CO2 are projected (e.g., GCAM Latin America B100 and B200, GRAPE Reforming Economies B100 to B300, IMAGE Asia B100 and B200, NLU multiple regions for B400). A combination of factors are contributing to this outcome, starting with the biomass feedstocks modeled and relative land use and land intensification opportunities that can result in changes in the competitiveness of a region in global agricultural commodity markets. For instance, a model converting less land in supplying biomass, and therefore resulting in smaller increase in land CO2, may be intensifying agricultural land management. We see evidence of this in the land nitrous oxide (N2O) results for GRAPE and NLU (see below).
Of the models that reported land N2O emission results (n = 7), most exhibit increasing land N2O emissions globally with increasing supply of modern biomass. However, like with land CO2, there are significant differences in the amounts emitted for a given level of biomass supplied, with the land N2O emissions from GRAPE significantly higher than that from the other models, and the GRAPE N2O emissions primarily from Asia and the Middle East and Africa regions. We also find variations in the sensitivity of land N2O emissions and their regional composition to the level of biomass supply. As with CO2, the N2O results are indicative of the feedstock types modeled, cost-effectiveness of feedstocks, and characteristics of the locations supplying the biomass. Note that, two models project decreases in land-based N2O emissions with increasing biomass supply (GCAM and GLOBIOM). These results are indicative of decreases in agricultural land cover acreage and livestock production, which offset agricultural production intensification on remaining croplands.
Underlying the differences in emission results between models are differences in land cover change (Fig. 3). All models exhibit increases in annual land cover conversions over time, and across scenarios, when there are increasing biomass supply requirements, but with notable differences in the specific lands converted. All of the models increase energy crop land cover, with three also increasing managed forest land (BET, GLOBIOM, NLU) and some showing more modest increases in other natural and non-energy cropland. The models vary significantly in terms of what types of land are being converted to support the increasing land cover types, with most models converting some amount of other (unmanaged) forest land, and some also converting pasture, non-energy cropland, and other natural lands (see Table 2 for model specific assumptions on allowed land conversions).
Among other things, the land cover changes inform our understanding of the models with modest land CO2 results in Fig. 2. For FARM, demand for forest products (wood and paper) with a larger and wealthier population is keeping managed forest land cover relatively stable over time despite growing demand for energy and food crops. For NLU, energy crops are only permitted on cropland and pasture, and energy croplands are assumed to have below ground carbon similar to pasture. For DNE21 + , while the model projects land-use change from forest and grassland to energy crops, the carbon stock (and CO2 emission changes) associated with the land conversions is projected to be modest in part because soil carbon changes are not modeled.
Non-energy crop price changes are an indicator of the opportunity cost of supplying biomass, i.e., the marginal cost. We are not able to directly compare biomass prices between models due to differences in model structure that impact how biomass prices are formed. In general, however, we find rising non-energy crop prices across models with increasing biomass supply—an indication of the rising cost of supplying biomass. The price impacts are modest for some models, but significant for others. Also, looking across biomass supply scenarios, we find some models exhibiting much larger increases in the non-energy crop price changes as biomass supply increases (MAgPIE and NLU). In these models, non-energy crop prices are more sensitive to the quantity supplied, indicative of inelastic supply, while those models that are less sensitive can increase the biomass supply with only modest biomass cost implications. We do, however, find one model with negative price changes (GCAM, except in the B400 case in 2100). This is indicative of the model’s modest land use change, and land productivity improvements outpacing biomass supply growth. Later in the paper, we are able to use biomass price experiments to explicitly map out biomass supply curves for some models.
Model specific biomass supply narratives
Looking across the different characteristics of supplying biomass, we tease out biomass supply narratives for each model. Table 3 presents summary metrics for each model’s B300 biomass supply results. Shown are global weighted average annual per unit energy feedstock, land cover, and emission results, as well as the non-energy crop price change results for 2050 and 2100, all relative to each model’s baseline scenario.
From these results, we can tell high-level biomass supply “stories” about each model’s tendencies and highlight differences between models. For instance, on average, biomass supply from AIM is primarily energy crops, with an increase in cropland acreage and decrease in forest and other natural lands resulting in an increase in land-based GHG emissions, as well as non-energy crop prices. Alternatively, the feedstock supply from GLOBIOM is more diversified, with residues and managed forest feedstocks representing about 50% of supply, with cropland gains similar to AIM, little loss of forest land, and a larger loss of other natural lands. The resulting global CO2 emissions are larger on average for GLOBIOM, while N2O emissions decrease and non-energy crop price changes are similar to AIM’s.
Across models, we see the largest annual average CO2 emissions from MAgPIE, IMAGE, and BET, driven by different average feedstock mixes and land cover changes, while the smallest are from DNE21 + , FARM, and NLU. For N2O, GRAPE produces the largest annual average emissions, while GLOBIOM exhibits an annual average reduction. Finally, for non-energy crop prices, NLU and MAgPIE exhibit the largest annual average increases, while the other models produce modest price increases.Footnote 4
While Table 3 is instructive for model-specific narratives and model comparison, it is difficult to derive generalizations, such as whether models relying solely on energy crops result in more land conversion and emissions. Underlying the different narratives, regarding for instance land CO2 emissions, are differences in assumptions such as soil and vegetation carbon and land productivity data, where there are large known uncertainties (Gasser et al 2020), which combine with each model’s intrinsic uncertainties in the representation of processes and the results produced.
Managing biomass supply land use and emissions externalities
The results thus far illustrate the potential for negative externalities from supplying biomass, with the potential for land conversion and emissions. Our land protection and mitigation sensitivity scenarios allow us to evaluate the implications of trying to manage these effects. Land constraints or pricing of emissions represent potential additional costs for producing biomass for energy. They also change the relative cost of feedstock types and locations. Costs will increase the most for feedstocks that are associated with greater land use change, especially on unmanaged lands, and/or more land emissions.
First, we find that the land GHG mitigation incentives result in a re-distribution of biomass supply, with changes in supply location and feedstock type. Figure 4 provides results for the 300 EJ/year in 2100 scenarios with land GHG mitigation versus without (i.e., B300C minus B300). A number of models shift supply out of Asia (e.g., DNE21 + , GRAPE, MAgPIE, AIM), and a number of models shift supply to the OECD (e.g., BET, FARM, GCAM, GRAPE, IMAGE). As for feedstock types, we see a few models with notable shifts away from energy crops towards residues (GRAPE, GCAM), but primarily we see shifts in the supplying location of energy crops and managed forest feedstocks. Finally, some models are much more sensitive to the land mitigation incentive (e.g., IMAGE redistributes over 50% of the supply).
Feedstock supply distributional change results for land protection and combined land protection and GHG mitigation are found in the SM (Figure SM2). We observe that some models have more feedstock substitution with land protection than they do with the GHG mitigation incentive (AIM, DNE21 + , FARM, and NLU), while others show greater feedstock substitution with the GHG mitigation incentive (GCAM, GLOBIOM, GRAPE, and MAgPIE). Recall, of course, that the default land protection assumptions vary by model (Table 2). We find that land protection shifts biomass supply away from Latin America in three of the models (AIM, IMAGE, NLU), but they differ on where the offsetting increases in biomass supply occur. Meanwhile, DNE21 + reallocates biomass away from Middle East and Africa and Asia, while FARM reallocates away from Reforming Economies. Interestingly, with land protections included, there is little movement away from energy crops and forest feedstocks towards residues. Primarily, the location of energy crop supplies is affected. Overall, land protection is resulting in a different feedstock distribution, as some lands are no longer available for use of any kind—biomass for energy, non-energy agriculture, and GHG mitigation. Note that, IMAGE and NLU were not able to meet the B300 annual supply requirements with their default land protection assumptions. Thus, these assumptions put an upper limit of their supply of biomass and deployment of bioenergy.
The effects on emissions and prices are, of course, of particular interest. Globally, we find that all models, except FARM and NLU, exhibit a land CO2 emission reduction, or even net uptake of carbon, relative to their baseline over the century when supplying biomass with the land GHG mitigation incentive (Fig. 5). Figure 5 provides results for the 300 EJ/year in 2100 scenarios with land GHG mitigation (B300C), land protection (B300LP), and both (B300CLP), as well as the pure supply result (B300). With the mitigation incentive, there are some increases in regional cumulative land CO2 emissions, but they are modest compared to the land carbon stock gains elsewhere. All the models exhibit land carbon gains in Asia, Latin America, and the Middle East and Africa, while there is some variation in sign across models for Reforming Economies and the OECD. These results are consistent with the shifts in the feedstock distribution we observe when there is a land GHG mitigation incentive (Fig. 4). For GCAM, GRAPE, and GLOBIOM, the large cumulative land carbon gains are associated with large and rapid afforestation.
For all but one of the models (AIM), the model-specific land protection assumptions result in lower net land CO2 emissions than without the assumptions (B300LP vs B300), with NLU’s land protections leading to a net increase in land carbon stocks. For AIM, land protection is resulting in greater land conversion of unprotected (and lower productivity) lands and therefore greater net land CO2 emissions. The combination of land GHG mitigation and protection results (B300CLP) are in between the individual sensitivity results for most models. However, which effect dominates varies across models. For two models (GRAPE and IMAGE), mitigation and land protection combine for greater carbon uptake in the terrestrial system.
As for land N2O emissions, increasing biomass supply with land mitigation incentives results in reductions in global land-based N2O emissions in all but NLU. Emission reductions are due to N2O being priced, reduced agricultural land conversions, and the resulting shifts in the distribution of biomass supplies. All the models exhibit land N2O emission reductions in Asia, Latin America, and the Middle East and Africa. These results are consistent with the shifts in the feedstock distribution we observe due to the land GHG mitigation incentive. In general, unlike CO2 emissions, we find that land protection assumptions have little or no effect on global N2O emissions.
Finally, regarding non-energy crop prices, we find land mitigation incentives and land protection assumptions resulting in significantly larger price increases than found with the pure biomass supply scenario (B300). Nearly all the models estimate larger price increases with land mitigation or land protection. Note that price increases are larger in 2100 than 2050 due to the increasing biomass supply over time, as well as the rising GHG price and growing demand for agriculture commodities due to economic and population growth. In general, we find that modeled land protections have a more modest affect on prices than the GHG mitigation incentive; however, together they result in higher prices than either does individually.
Global and regional estimated biomass supply curves
Five modeling teams have also run biomass supply curve scenarios driven by exogenously specified prices for modern biomass (AIM, FARM, GLOBIOM, IMAGE, MAgPIE). These scenarios reveal the economic supply curves for biomass implicit within each model. With a set of increasing biomass farmgate price experiments of $3, $5, $9, and $15 per gigajoule (US$2005/GJ), we tease out each model’s implied supply curves. We implement each biomass price globally and in all time periods (i.e., identical in all regions and across time). For models not able to impose a farmgate price, a delivered marginal wholesale biomass price is used, where the point of delivery is the theoretical edge of each regional energy system.
From the five models, we find upward sloping global biomass supply curves, with the quantities supplied increasing with the price (Fig. 6). However, we also find differences across models in the location and slope of the curves, with FARM suggesting the greatest supply at any price, and AIM or GLOBIOM the smallest supply. AIM’s supply is the most inelastic (i.e., least responsive to price) and FARM’s the most elastic, with differences in land allocation formulations between models likely contributing to these differences. For instance, FARM allocates land according to relative returns, while other models use other mechanisms such as logit, constant-elasticity-of-transformation formulations, and exogenous prioritization (see Table 2 for details).
At a given level of demand (i.e., price), we find biomass supply curves shifting out over time, indicating declining opportunity costs due, in large part, to assumed technological improvements in land sector productivity despite rising food, feed, and wood product demands. The supply curves for FARM and IMAGE shift out the most over time, while AIM and GLOBIOM shift out the least.
Within each model, we find substantial differences in biomass feedstock supplies across regions. While across models, we find large differences in implied supply for a given region, as well as differences in the ordinal ranking of regional supply, with four models estimating the largest potential supplies from Latin America and the smallest from Reforming Economies (AIM, GLOBIOM, IMAGE, MAgPIE), and another model estimating the OECD and Reforming Economies to have the largest supplies (FARM). The differences in estimated global and regional feedstock supplies between models are due to numerous differences in models, including land availability, productivity, feedstock types, and land and commodity markets. Exploring the Fig. 6 regional results further (Table SM2), we find that the models notable differ in energy crop yields (in gigajoules per hectare) and energy crop share of supply, with higher regional yields and shares associated with higher energy crop use (e.g., FARM with higher yields in Latin America versus GLOBIOM, or FARM's yields in the OECD versus AIM as well as the other models). However, yields alone are unable to fully explain regional supply differences between models. Supply differences are also influenced by, among other things, differences in production costs, commodity prices and opportunity costs, and modeling structure (e.g., feedstock options and specifications, land eligibility, and land conversion possibilities).
It is important to keep in mind that these feedstock supply estimates are supply curves, capturing the production and opportunity costs and energy value of feedstocks. The GHG emission implications of producing and using the feedstocks are not priced (or constrained) in these estimates of supply potential, but would be captured in integrated modeling solutions when land emissions can be explicitly or implicitly priced, land mitigation incentivized, and/or feedstock and land constraints activated. To help us think about the implications of these integrated elements on biomass supply, we also run the set of price experiments with the land GHG mitigation incentive. In Figure SM3, we clearly see that incentives for land-based mitigation increase the cost of providing biomass, shifting the biomass supply curves inward. Also, consistent with our earlier observations regarding the supply impact of land mitigation, the sensitivity of biomass supply to the land GHG mitigation incentive varies across models and over time with small (AIM) and large (IMAGE) shifts in supply curves implied.
Overall, the supply curve representation of biomass supply provides another level of understanding, informing interpretation of the earlier feedstock supply distribution results (Fig. 1), as well as integrated modeling use of biomass (next section). Among other things, the supply curves make explicit the least-cost regional ranking within each model and shed light on which models perceive biomass as less or more expensive as a fuel.
Integrated biomass quantity demanded
The results above elucidate key aspects of supplying biomass, which are necessary for evaluating the long-run opportunities for bioenergy in climate management—potential least-cost feedstock supply, and potential GHG emissions, food price, and land conversion implications of supplying increasing amounts of biomass for energy. As such, they inform our interpretation of integrated modeling results, where the supply and demand for biomass (and associated land and conversion GHG emissions) are modeled simultaneously. In integrated modeling, the biomass demand is a function of the markets for providing economic goods and services, as well as decarbonization, and the resulting biomass quantities represent the quantity demanded in market solutions. We therefore conclude our results section with an evaluation of integrated biomass quantity demanded results in terms of our insights from the biomass supply experiments. Here we take advantage of the EMF-33 “bioenergy demand” experiments (Bauer et al., this issue), which provide projections for potential future energy and land systems for a scenario consistent with global climate goals. These results include projected deployments of bioenergy technologies and the biomass feedstock quantities supplied, as well as deployment projections for non-bioenergy technologies and fuels. In identifying potential bioenergy strategies in the context of global decarbonization for future climate management, the integrated modeling solutions account for all of the biomass supply features we have revealed above, as well as other factors.
Figure 7 provides the regional and feedstock type market solutions for quantities of biomass demanded over time from the EMF-33 integrated modeling climate management scenario consistent with limiting global average warming to 2 °C. This particular scenario included model default full mitigation technology portfolios, including bioenergy with CCS (BECCS), and endogenous modeling of land-based emissions and carbon pools.Footnote 5 The first thing to note is that the models clearly project that it is cost-effective to use increasing amounts of biomass globally over time. However, annual biomass use and growth vary significantly between models for the same climate future, with 2100 total global biomass use reaching as little as 87 EJ/year (IMACLIM-NLU) and as much as 424 EJ/year (REMIND-MAgPIE). We also find that residues and energy crops dominate the market supply quantities of all the models, and managed forest feedstock use is notably absent from models that include it as an option (BET, FARM, MESSAGE-GLOBIOM, GRAPE, REMIND-MAgPIE), except for a visible amount in FARM early on.Footnote 6 In general, as discussed below, the price on land emissions is high, creating a strong incentive to store carbon in existing and new forests and increasing the cost of supplying forest and other biomass for energy.
In addition, we find that the regional (and feedstock) distributions differ sharply from the least-cost supply distributions (Fig. 1), which illustrates the role of the other biomass supply factors that are accounted for in integrated modeling, such as the estimated land GHG emissions of supplying biomass, the biomass supply implications of pricing land emissions, and the assumed land protections (Table 2) and their implications for supplying biomass, as well as the market for bioenergy, which depends on energy technology options. Every model, except MESSAGE-GLOBIOM, has a notably different biomass supply distribution across regions compared to Fig. 1 (for similar biomass quantity levels). For instance, we observe a significant shift away from supplying biomass from the ME/Africa in AIM, while FARM shifts away from biomass from Reforming Economies towards supply from the OECD, Asia, and ME/Africa, and IMAGE shifts from a supply dominated by Latin America towards one more balanced across regions.
As noted, the EMF-33 integrated scenarios price land GHG emissions, which we saw above can have a significant impact on feedstock supply as well as land emissions (Figs. 4 and 5). In this case, however, the integrated modeling GHG prices are much higher than the GHG price path we applied in our biomass supply scenarios, ranging from $104 to $1,260/tCO2 in 2050, versus $49/tCO2 in 2050 from our biomass supply land mitigation sensitivity scenario. See Table SM3 for a summary of additional integrated modeling results—global biomass primary energy in 2100, cumulative land CO2 and N2O emissions, 2050 GHG price, and 2050 change in non-energy crop prices. Regarding land emissions, we observe that in most models, the cumulative CO2 and N2O emissions are directionally similar, but the strength of the implications has changed. Some models, however, do exhibit changes in the sign of the affect. For instance, taking into account differences in biomass supply quantities, we find GRAPE and IMAGE producing increases in land CO2 emissions (Table SM3) versus decreases found in their biomass supply only modeling with a carbon price and land protection (Fig. 5), and we see changes in the sign of N2O emission effects for MESSAGE-GLOBIOM and REMIND-MAgPIE.
With many moving parts, integrated results and differences across models are difficult to unpack. However, our detailed biomass supply assessment helps us understand the integrated outcomes. For instance, we found earlier that supplying biomass is relatively more expensive for NLU than the other models in terms of non-energy crop price impacts, which contributes to IMACLIM-NLU’s relatively limited use of biomass in Fig. 7. Similarly, in the integrated scenario, AIM uses relatively little biomass due in part to is inelastic and relatively expensive biomass supply (Fig. 6), while IMAGE uses relatively little biomass due to a combination of factors, including the model allocating land for food production first, which precludes biomass feedstock production competition with food, as well as its large land CO2 emissions associated with supplying biomass and the significant decrease in its biomass supply curve when land emission are priced. Alternatively, with FARM’s biomass supply relatively inexpensive and only modest land CO2 emission implications, FARM uses a large amount of biomass in the integrated scenario. While GCAM and MESSAGE-GLOBIOM use more moderate amounts of biomass in the integrated scenario, they do so with large cumulative net land CO2 uptake due to significant afforestation, which, in GCAM, results in an over 500% increase in 2050 non-energy crop prices.Footnote 7
While the biomass supply insights inform our understanding of integrated modeling results, there are still additional factors in play that cause the integrated outcomes to differ from what the individual model biomass supply results would suggest on their own. For example, the biomass supply characteristics we have elucidated do not give us the full story for REMIND-MAgPIE. REMIND-MAgPIE uses a large amount of biomass in the integrated result, but biomass supply in MAgPIE is associated with relatively high land CO2 emissions and non-energy crop price impacts, with the price impacts even greater with land GHG mitigation incentives and land protection assumptions. In addition to considering the costs and emissions associated with supplying different regional levels and types of biomass, as well as land protection constraints, the integrated modeling endogenously models the demand for biomass, which represents bioenergy technology cost and performance assumptions and feedstock preferences (Daioglou et al., this issue), bioenergy local consumption constraints and trade opportunities, and the cost-effectiveness of bioenergy relative to other decarbonization options across the global economy and over time.
Finally, and importantly, the integrated results highlight that, even when considering land emissions, land conversion, and crop price implications, bioenergy could still be a large and cost-effective long-run climate management strategy. Thus, employing bioenergy as a part of the portfolio for managing the climate system will likely have trade-offs, and those trade-offs could be cost-effective in terms of minimizing the net decarbonization welfare implications for society.