Modeling meets science and technology: an introduction to a special issue on negative emissions

2.1 Models, targets, and strategies

The four IAMs are century-scale models that either simulate (GCAM and IMAGE) or optimize (REMIND and WITCH) a cost-effective strategy to reduce greenhouse gas emissions. In the simulation models equilibria are developed for each point in time, while for the optimization models the whole temporal dynamics is resolved at once, assuming a forward-looking planner with perfect foresight who discounts future wellbeing.

In all cases the target is a concentration of CO_2e not to be exceeded in 2100. The target concentration is not exactly equal across papers, but it is stringent and roughly in line with an increase of the average surface temperature capped at 2.0 °C to 2.5 °C. Importantly, the concentration objective can be exceeded at intermediate times, because subsequently CDR can reduce the CO₂ concentration.

Climate feedbacks on these trajectories are not included in the models. Thus, incremental damage during overshoot periods is not taken into account. Damage-trajectory relationships are complex, because damage is associated with both short-term forcing (which impacts meteorology) and integrated forcing (which affects large-time-constant behavior such as icesheet melting and ocean warming). With a slightly lower concentration in 2100 than in 2070, for example, should one expect to see any reduction by 2100 of the storm severity and ice melting seen in 2070?

The technological resolution of CDR varies across the models. Edmonds et al. and Van Vuuren et al. model “yield” (biomass production per hectare per year) and land area. Only Edmonds et al. develops the biomass energy supply from a model (GCAM) that fully couples the economy, energy, and land use. The other models rely on a soft coupling to the land and introduce cost curves for bioenergy supply exogenously.

As summarized in Table 1, four CDR strategies are explored by at least one model. All models give prominence to biological energy production with CO₂ capture and storage (BECCS). Edmonds et al. also includes afforestation and storage of carbon in materials (such as plastics). Both Chen & Tavoni and Fuss et al. include Direct Air Capture of CO₂ with chemicals (DAC). These four strategies illuminate many of the key issues raised by CDR, notably co-benefits, gross vs. net emissions, and demands on land. Among the CDR strategies that are not explored at all are ocean-based strategies such as iron fertilization and land-based strategies such as biochar.

From the perspective of integrated assessment modeling, the CDR options are just another set of technologies for mitigating climate change, competing with the usual list of strategies, such as efficiency, renewables, CCS, and nuclear power. Cost trajectories and quantity constraints for the CDR strategies are the key new information introduced into the models. This competition is particularly transparent in Fuss et al., where the assumed cost of removing a marginal percent of baseline CO₂ emissions is constant over time but depends exponentially on the percent already removed: the marginal costs of abating 25 %, 50 %, and 75 % of baseline emissions are, respectively, about $10/tCO₂, $100/tCO₂, and $1000/tCO₂.

The four IAMs disaggregate the world into macro-regions (roughly 12–14 macro regions for all models except IMAGE, which has twice as many) which interact through trading. Edmonds et al. and Chen & Tavoni explore both differential regional compliance with global agreements and differential regional endowments affecting CDR capability (such as geological storage capacity). In Edmonds et al. and Chen & Tavoni, delayed participation can result in regional emissions that exceed some pre-specified cumulative global emissions, making the corresponding CO_2e concentration target unobtainable. As a consequence of differential endowments, Chen & Tavoni expects BECCS to be implemented primarily in Latin America and assigns DAC especially to the transition economies and the MENA region (Middle East and North Africa). Edmonds et al. sees China, the U.S., and Latin America dominating, when CDR includes both afforestation and BECCS.

2.2 Biological and geological constraints on CDR

All biological strategies in these essays are terrestrial: afforestation and reduced deforestation, biopower without CCS, biopower with CCS (BECCS), and biofuels. They all eventually encounter land constraints. They compete for land with each other and with uses of land for food and fodder, non-food products from forestry and agriculture, recreation, and wilderness preservation.

Limited geological storage capacity for CO₂ produces another constraint. The total global CO₂ storage capacity, assuming minimal problems with public acceptance, ranges from 2100 GtCO₂ in Fuss et al. to 7000 GtCO₂ in Edmonds et al.—in all cases ample for century-scale conventional fossil-fuel-based CCS, BECCS, and DAC. However, regional storage constraints do arise and give regions with large storage potential a competitive advantage in conducting CDR, provided inter-regional CO₂ transport is absent. Inter-regional CO₂ trade is subjected to a sensitivity analysis in Chen & Tavoni.

2.3 Resulting level of deployment

Table 1 shows that cumulative CDR for the 21st century is on the order of 1000 GtCO₂, bracketed by 500 GtCO₂ and 1600 GtCO₂. By comparison, currently, the annual emissions rate from fossil fuels is 30 GtCO₂/yr and the atmosphere contains 3000 GtCO₂. In Edmonds et al. the ranges for estimated 21st century CDR from afforestation and BECCS overlap; similarly, Chen & Tavoni allows both BECCS and DAC and finds comparable contributions. Cumulative CDR activity represents a substantial fraction of total estimated mitigation over the century, roughly 3500 to 4000 GtCO_2e for stringent climate targets. A CDR contribution of 1000 GtCO₂ reduces the equilibrium average surface temperature about 0.5 °C, according to the carbon budget approach (Matthews et al. 2012).

The peak rate of CDR activity is similar across all five models, ranging from 10 to 35 GtCO₂/yr. In Edmonds et al., but not in the other four IAM papers, CDR enters significantly early in the century. It enters in the form of afforestation, which quickly becomes competitive with other mitigation options, because terrestrial carbon emissions are penalized and rewarded at the same carbon price as fossil-fuel carbon emissions. The result is early, large, and widespread changes in land-use practices.

In summary, the IAM models show that when strong targets for carbon mitigation are pursued, CDR technologies costing between $100/tCO₂ and $1000/tCO₂ contribute substantially to climate-change mitigation in the second half of the century. Some high-cost mitigation options are never deployed. In the final period of the 21st century, for the reference case in Fuss et al., one-third of baseline emissions remain unmitigated. In effect, residual emissions enter the atmosphere at one location, and at other locations a net-cumulative-emissions target is met by CDR that removes nearly an equivalent amount of CO₂.

Overshoots are reported for three of the five models. They are much smaller in Kriegler et al. and Chen & Tavoni (40 ppm and 60 ppm for concentration overshoot, 0.05 °C and 0.06 °C for temperature overshoot, respectively) than in Edmonds et al. (where the overshoot ranges are 55 ppm to 135 ppm and 0.2 °C to 0.5 °C). The disparities in the ratio of temperature overshoot to concentration overshoot across these three papers reflect differences in the timing of CDR implementation and the lagged relationship between concentration and temperature.

As seen in Table 2, in 2100 the bioenergy contribution to global primary energy ranges from 100 to 300 EJ/yr, across the five papers. By comparison the models estimate roughly 1000 EJ/yr of total global primary energy at that time. Not all of the bioenergy will appear as BECCS, since some will be for biofuels and some will be for biopower without CCS. Table 2 also reports land requirements of several hundred million hectares (Mha) for biological CDR for the two IAM models that include land explicitly: Edmonds et al. reports 490 Mha and 530 Mha for two scenarios where land is assigned to both afforestation and BECCS, while van Vuuren et al. estimates from 300 Mha to 600 Mha, all for BECCS.

BECCS is first implemented early in the century (around 2030), but most CDR via BECCS is achieved after 2050, when deep cuts in emissions are needed and carbon prices exceed $100/tCO₂. However, BECCS is adversely affected by the rising cost of biomass that reflects its constrained supply. As seen in Table 2, biomass prices in 2100 for three baseline cases without BECCS are between $3/GJ and 8$/GJ, similar to the models’ prices today. But when BECCS is available for the same targets, it produces dramatic upward pressure on the biomass price, which increases to 31–94 $/GJ, depending on marginal production costs and scarcity rents.

2.4 Implications for reduced effort on conventional mitigation

In all five IAM models, CDR availability reduces conventional mitigation early in the century, relative to base cases where no CDR is available. This is expected: in any cost optimization over several time periods, knowing for sure that an extra option will be available at a future time reduces the effort expended on every other option at earlier times—relative to the effort to meet the same goal when there is no extra option. The extent of late CDR and associated reduction in early conventional effort depends on the discount rate. The discount rate captures the willingness to transfer mitigation costs to future generations that will also face higher climate damages. Lowering the discount rate indeed reduces the scope for CDR in Chen & Tavoni. Given the potency for policy making of the issue of appropriate current effort, the sensitivity of specific results to key assumptions in the IAM papers should make the reader cautious about carrying modeling results into the real world.

3.1 Direct air capture with chemicals (Mazzotti et al.)

Mazzotti et al. presents an optimization of the same benchmark system that was studied, but not optimized, in the report on DAC released in 2011 by the American Physical Society (Socolow et al. 2011). Optimization across three variables (fraction of CO₂ captured, air flow velocity, and sorbent solution velocity) reduces the estimated capture cost only slightly, because key cost assumptions for capital equipment are not changed. The cost estimate (in the range of $500 per ton of CO₂) is used as an input in the integrated assessment papers by Fuss et al. and Chen & Tavoni.

The APS report and Mazzotti et al. propose that CO₂ removal from the atmosphere will almost certainly be substantially more expensive than reducing (by retrofit or rebuild) nearly all of the emissions from large and concentrated sources of CO₂ at coal and natural gas power plants, refineries, cement plants, and many other industrial facilities. The fundamental reason is that the concentration of CO₂ in the flue gas at these facilities is much higher than in air. The ratio for a coal power plant is roughly 300–12 % CO₂ in the flue gas, 0.04 % in air. This ratio affects the frontal area facing the incoming gas, but this is only one of several contributors to the cost ratio. The path of flue gas at a coal plant can be longer, the percent of CO₂ removed can be larger, materials can be allowed to be more expensive, and operating costs can be a higher fraction of total costs. In Mazzotti et al., the ratio of total costs is 6 to 8.

Several start-up companies are seeking to reduce the cost ratio further. Lackner, for example, speculates that DAC could take the form of large numbers of low-unit-cost factory-produced absorbers. (Lackner et al. 2012) Arguably, DAC could experience greater cost reductions than flue-gas capture, because a DAC facility has only a single objective, whereas conventional CCS at a large industrial facility is an add-on, complicating and compromising the core activity (e.g., producing power). Nonetheless, it is hard to imagine a future where many coal power plants spew their CO₂ into the atmosphere while, simultaneously, many DAC plants are removing CO₂ from the atmosphere. It seems far more likely that DAC deployment will not begin in earnest until very few large concentrated sources remain.

Perhaps DAC can expand greatly with little change in unit cost, making DAC a “backstop” technology. By contrast, BECCS will eventually reflect land constraints. Moreover, DAC may require much less land per ton of CDR than BECCS and may create fewer land-use conflicts. In common, BECCS and DAC require low-cost, publicly acceptable, and abundant geological storage capacity for CO₂.

3.2 Ocean outgassing (Vichi et al.)

CO₂ is removed from the atmosphere by natural processes that restore equilibrium. Extra CO₂ in the atmosphere leads to some fraction dissolving into seawater at the ocean’s surface, and a similar redistribution (though much more complicated) occurs for vegetation. Over the past several decades, on average (but with large year-to-year variation) the net atmospheric removal rate has been about half of the emissions rate from fossil fuel combustion (with roughly equal contributions from ocean and land removal). One can anticipate, therefore, that deliberate removal of CO₂ from the atmosphere should be accompanied by ocean outgassing of CO₂ and (with less certainty) reduction in the global stock of carbon in biomass. These are negative feedbacks on the CDR enterprise. Vichi et al. may be the first paper to estimate ocean outgassing when the CO₂ is removed over an extended period of time. Cao and Caldeira estimated ocean outgassing when CO₂ is removed from the atmosphere at a single moment (Cao and Caldeira 2008).

Vichi et al. studies two related extended campaigns that in the absence of outgassing would remove about 490 GtCO₂ (130 GtC) from the atmosphere over a 30 year period (about 16 GtCO₂/yr, about half the current rate of fossil fuel emissions). As discussed in the previous section, CO₂ removal rates this large and larger are foreseen by the IAMs. In the first campaign, a global CDR industry follows a prescribed emission-reduction schedule: it removes 16 GtCO₂/yr for 30 years and ignores the outgassing; in response 50 GtCO₂ outgases from the ocean, a 10 % negative feedback. In the second campaign, a global CDR industry follows assures that the net CO₂ removal is 16 GtCO₂/yr when outgassing is taken into account. This requires removing somewhat more than 16 GtCO₂/yr, because the concentration difference between ocean and atmosphere is always larger; the associated outgassing is slightly greater, 66 GtCO₂. The authors warn that their results are specific to an initial atmospheric concentration of CO₂ of 348 ppm (the concentration in 1988) and for a specific ocean model. Nonetheless, a crude representation of outgassing in Chen & Tavoni indicates that this negative feedback can significantly reduce the scope for CDR.

Although the terrestrial sink is much less well understood than the ocean sink, by a combination of measurement and modeling a coarse sketch of the geographical distribution of the terrestrial carbon sink is emerging. Eventually, models will incorporate a CDR-efficacy factor that accounts for all sources and sinks. For now, a conservative estimate—aligned with the strength of the negative feedbacks from the ocean and land observed today—would assume that a one-ton reduction in the atmospheric stock of CO₂ requires two tons of CO₂ to be removed from the atmosphere.

3.3 Land, nutrients, and water for biological CDR (Smith & Torn)

Smith & Torn studies the environmental aspects of two biological CDR strategies: BECCS and afforestation. Their BECCS system is based on switchgrass grown in a temperate climate. Afforestation is modeled as 50 years of carbon build-up in tropical eucalyptus plantations. In both cases, these choices are intended only to be illustrative: to illuminate qualitative issues common to many biomass strategies and to provide templates for analogous calculations bearing on other species and locations. They estimate demands on land, water, nitrogen, and phosphorus and they identify impacts on albedo, biodiversity, and land tenure. All quantitative estimates are scaled, helpfully, to a global enterprise removing carbon dioxide from the atmosphere at a rate of 3.7 GtCO₂/yr (1 GtC/yr). Here, we report a few of their sobering results.

The key estimate is the productivity of land, i.e., the yield in tons per hectare-year (t/ha-yr). For the Smith & Torn switchgrass plantation producing feedstock for BECCS, the estimated yield is 5.6 to 23 t_biomass/ha-yr. Note that the high estimate is four times larger than the low estimate. Assuming that the carbon intensity of energy from switchgrass is 15 GJ/t_biomass, the estimated yield of primary energy in the form of switchgrass feedstock (rounding off) is 80 to 350 GJ/ha-yr. Compare this range with the primary biomass energy yield at the end of the century assumed for two scenarios in Edmonds et al., 600 GJ/ha-yr (about 300 EJ/yr grown on about 500 Mha, see Table 2). A rate of improvement of yields of 0.6 %/yr to 2.3 %/yr for 90 years would be required for Smith & Torn’s estimated yield today to grow to the yield in 2100 in Edmonds et al. This range brackets the 1.6 %/yr long-run record of yield improvement reported by the Food and Agriculture Organization of the United Nations.

The “net carbon” issues for DAC explored in Mazzotti et al. reappear in the BECCS system examined by Smith & Torn. The authors estimate that about 2.2 tons of carbon must be fixed in switchgrass by photosynthesis for each ton of carbon sequestered as CO₂ in a geological formation via BECCS, to compensate for emissions associated with harvesting, transport, and electricity production. Smith and Torn report that switchgrass is 42 % carbon; as a result, to sequester one ton of carbon per year via BECCS will require growing 5.2 tons of biomass per year, corresponding to 0.23 to 0.94 hectares.

As for land demands for afforestation at eucalyptus plantations, Smith & Torn assume that carbon builds up in trees for 50 years at an average rate of 1.3 to 4.0 tC/ha-yr. Thus, each ton per year of CDR via afforestation will require 0.25 to 0.77 hectares. Evidently, in Smith & Torn, the land required for an identical level of CDR activity is nearly the same for a fifty-year afforestation project and for BECCS. At the end of the 50 years, new land will be needed for afforestation (the eucalyptus trees will have reached maximum size), but, in principle, the land planted in switchgrass for BECCS can keep producing.

If hundreds of millions of hectares are dedicated to carbon management, other land will experience knock-on effects. For example, a change from pasture to energy crop for BECCS will require that fodder be grown elsewhere and may elicit feedlot cattle-raising. These knock-on consequences are subsumed under the label, “indirect land-use change,” or “ILUC” (Searchinger et al. 2008; Fargione et al. 2008). They will have their own climate impacts—not only CO₂ emissions but also changes in albedo and evapotranspiration. Including ILUC in the assessments of BECCS and afforestation was beyond the scope of Smith & Torn.

Smith & Torn estimates that an average application rate of 80 kgN/ha-yr of nitrogen fertilizer is required to achieve the assumed switchgrass yields. The Smith & Torn reference CDR level of 1 GtC/yr (3.7 GtCO₂/yr), therefore, will require an annual input of 19 to 75 million tons of nitrogen. By comparison, today’s annual global fertilizer production provides about 100 million tons of nitrogen. Today’s rate also roughly equals the global rate of biological fixation of atmospheric nitrogen by microorganisms. Increased nitrogen availability has fed the world, but it has also had many adverse environmental consequences, such as eutrophication of coastal waterways and increased production of nitrous oxide (N₂O), a potent greenhouse gas. BECCS can be expected to exacerbate these adverse consequences, even as it retards climate change. The authors speculate, however, that future nitrogen requirements to achieve the same yield will decline as fertilizer is delivered more effectively.

Hundreds of millions of hectares dedicated to biomass CDR is an immense demand on land. Integrated assessment models that call for large demands on land have long incorporated ecological constraints that reduce the scope for bio-energy in climate change mitigation (van Vuuren et al. 2009). It is already time for broad discussion of these ecological constraints and their implications. Climate mitigation modelers, land use planners, agricultural economists, agronomists, foresters, plant physiologists, and genetic engineers—this is only a partial list of those who must participate.

4.1 Pace of change

Every attempt to describe how the future differs from the present must deal with the question of how quickly change can occur—what we call the “pace of change” (or, simply, “pace”). Integrated assessment modeling, for example, embeds constraints that prevent one component of the energy system, such as nuclear power, from displacing all its competitors in a single decade, even if the model introduces a technological breakthrough that suddenly lowers this alternative’s cost. Necessarily, judgment is required to quantify constraints on pace. History can help by providing evidence bearing on the time required, for example, for one energy alternative (such as automobiles) to displace another (such as horses). Nothing happens overnight.

Pace for the biological CDR options is constrained by how quickly land use can change. In the reference case (T1) in Edmonds et al., a uniform price on biological and fossil fuel carbon of $16/tCO₂ ($60/tC) is imposed in 2015–2020, which “dramatically changes the economics of land use and dramatically changes the allocation of land in the model.” (See Edmonds et al., Supporting Online Material.) As a result, in that same initial time step, no less than one billion hectares of land “change use to forest systems predominantly, but not exclusively, in the tropics.” This is the land area of the United States. On average, each hectare undergoing afforestation is sequestering 4 tC/yr (15 tCO₂) by 2020, resulting in a new land sink of 15 GtCO₂/yr that cancels roughly half of the world’s fossil fuel emissions. The average “pace” of this rapid increase is 3 GtCO₂/yr². Bottlenecks that might constrain land use change on one billion hectares in a five-year period that begins just two years from now are not internal to this specific mitigation strategy in this specific model.

To take pace into account and penalize overly rapid deployment, Kriegler et al. introduces “adjustment costs” on the rate of capacity expansion. Van Vuuren et al. observes that the “emission reduction rate” (our “pace”) is constrained by the cost of “premature retirement” as well as “the inertia in the change of consumer preferences, international negotiation processes, policy formulation and the maximum deployment rate of new technologies… These factors are very uncertain and typically not captured in energy-economic models.” Meadowcroft, contemplating large-scale CDR quite generally, writes: “Scale must also be thought about in terms of the human activity required to realize the approach, the scope of existing societal practices that must be transformed, the land-use footprint and potential environmental impacts. And this suggests ambitious estimates of CDR potentials—especially by mid-century—must be viewed with caution.”

All CDR modeling will eventually take pace into account. For DAC, constraints on pace arise from limits on the rate of build-up, globally, of specific industrial activity, such as cement or steel production, and the fraction of such a build-up that could be dedicated to DAC. Historical precedents might be sought in the rates of build-up observed during wartime mobilization or, perhaps, in the rates of increase in specific industrial sectors (such as road and building construction) during the recent period of high economic growth in China.

The CDR pace can be quantified in the units of the “stabilization wedge” (Pacala and Socolow 2004; Socolow and Lam 2007). A wedge is defined by linear growth of a carbon-responsive strategy over 50 years, starting from no activity at all and rising to an activity level of 1 GtC/yr. A one-wedge pace, therefore, is 0.02 GtC/yr², or 0.073 GtCO₂/yr². Thus, the pace of afforestation discussed above, 3 GtCO₂/yr², is a 40-wedge pace. Constraints on pace are an important challenge to modeling in general and to the modeling of CDR in particular.

4.2 Matching CDR to threats

Yet another frontier of CDR research would scope out CDR’s capacity to respond to threats arising from climate change. Threats can be chronic or acute. A chronic threat could take the form of a “new normal” in climate that is widely judged to be unfavorable, relative to an earlier climate. An acute threat would be a “climate emergency”: the sudden appearance of a dangerous manifestation of climate change that threatens to overwhelm civilization.

To counter a chronic threat, CDR may be used someday to lower the atmospheric CO₂ concentration. At such a time nearly all fossil fuel emissions would have ended. How quickly could the CO₂ concentration be reduced? As summarized in Table 1, the papers in this issue contemplate CDR at 15 to 30 GtCO₂/yr by 2100. Conservatively, at this rate of deployment, the CO_2e concentration will fall by 1 to 2 ppm/yr (see above). Think 550 ppm being lowered to 450 ppm over 50 to 100 years. In short, CDR could be matched to a globally coordinated deliberative process that adjusts the atmospheric concentration on a century-scale.

CDR, however, is not matched to a climate emergency, which is likely to require that an effective response be mounted over a decade or so. Examples of emergencies are rapid melting of the Greenland icesheet and rapid outgassing of methane now sequestered in buried ice in structures known as clathrates. It is possible, however, that a sufficiently rapid response could be generated by Solar Radiation Management (SRM). The objective of SRM strategies is to increase the percent of incoming sunlight reflected by the planet (e.g., from its current value of 31 % to 32 %), thereby countering the warming by greenhouse gases. Strategies to increase the planet’s reflectivity include making clouds whiter and injecting reflective particles into the stratosphere. Often today CDR and SRM are presented as two versions of “geoengineering”—a word intended to encompass all forms of deliberate intervention in the earth’s physical, chemical, and biological systems at the planetary scale (Royal Society 2009).

4.3 Iterative decision-making

The world is certain to surprise us. In order to be responsive to the uncertainties that lie ahead, several IPCC assessments and a recent report from the National Academy of Sciences, America’s Climate Choices (National Research Council of the National Academies 2011) recommend “iterative risk management.”

Iterative risk management is essential for CDR, where uncertainty permeates every aspect. Consider three sources of uncertainty. First, uncertainty looms over the stabilization target, and, as seen in several of the papers in this issue, the extent of deployment of CDR is tightly coupled to the climate objective. In the modeling analysis in this special issue, the target-setting process is idealized as the setting of a global target very soon and its retention without modification throughout the century. This idealization is mathematically convenient, but it obviously does not capture reality. Modeling should be able to explore the robustness of early actions in a world where countries continually revise their climate change policies and targets and, through some collective process, iterate global targets (Haurie et al. 2011). Revisions in national policies and practices will reflect both the evolution of climate science and what nations learn from experience about the effectiveness of their own and other countries’ mitigation and adaptation efforts.

The competitiveness of fossil fuels is a second major source of uncertainty. Oversimplifying, there are two views today about how quickly the fossil-fuel era can be brought to a close: in just a few decades or in more like a full century. Underlying this difference in views are judgments not only about the urgency of dealing with climate change but also judgment about the competitiveness of fossil fuels (without CCS), relative to low-carbon alternatives (Socolow 2012).

A third uncertainty surrounds judgments about what Meadowcroft calls the “social impacts” of technology. Meadowcroft writes: “Typically it is only as a technology is rolled out into society that one can get a firm grip on the timing and strength of side effects, the operation of countervailing forces, and the mobilization of direct opposition.” For example, biological versions of CDR will undergo periodic assessment of impacts on land tenure and property rights that could affect level of deployment.

Uncertainties about targets, fossil-fuel competitiveness, and social impacts of CDR are obviously linked. Analysis on behalf of iterative decision-making regarding the value of CDR will require contributions from practitioners of disciplines only minimally represented in the articles in this issue, including policy analysts, political scientists, lawyers, economists, psychologists, and sociologists.