(Mis)conceptions about modeling of negative emissions technologies

Intentionally removing carbon from the atmosphere with negative emission technologies (NETs) will be important to achieve net-zero emissions by the middle of the century and to limit global warming to 2 °C or even 1.5 °C (IPCC 2018, Minx et al 2018, Strefler et al 2018, van Vuuren et al 2018). However, to what extent and when the various NETs will be utilized to complement emission reductions in future climate policies is still unclear. Recent assessment reports provided for example by the Royal Society (2018) or the National Academies of Sciences, Engineering, and Medicine (2018) discuss a broad set of potential technologies while current process-based Integrated Assessment Model (IAM) analyses predominantly restrict their attention to reforestation, afforestation and bioenergy with carbon capture and storage (BECCS) as part of the climate policies. Earth System Model (ESM) analyses on the other hand, still predominantly focus on stylized, large-scale scenarios. In scenarios for afforestation or ocean alkalinity management it is assumed that the entire Australian hinterland will be planted, that global forest will increase by 8 million km² (all 'marginal' land), or that alkaline minerals are utilized on the order of (or many magnitudes beyond) the global coal industry mining activities (Keller et al 2014, Sonntag et al 2018).

Both, ESM- and IAM-based analyses consider so far specific technologies rather in isolation and not as part of a broad portfolio of NETs. However, the implementation of portfolios of NETs (also over time, given the different saturation levels of different NETs) is considered to increase the feasibility, reduce the side-effects, and in turn improve the sustainability of carbon dioxide removal in climate policies compared to the large-scale application of a single technology (Fuss et al 2013, IPCC 2018). Whether or not either type of model can adequately simulate NETs singly or as a portfolio, even in idealized scenarios, is an important question. Simulating the climate and carbon cycle today, and in the recent past, where we have observations and can assess model performance, is challenging and we know that the representation of many key processes needs to be improved. When it comes to simulating NETs these difficulties may be exacerbated, potentially in unknown ways as we have never observed how the Earth system responds to intentional CO₂ removal. Our current understanding suggests that intentionally removing carbon from the atmosphere by enhancing natural carbon sinks, engineering new carbon sinks, or combining natural uptake with engineered one would not linearly extend the carbon budget because in an interacting carbon cycle their net contribution will be strongly influenced by feedbacks and saturation effects (Fuss et al 2014, Tokarska and Zickfeld 2015, Jones et al 2016, Keller et al 2018). Accordingly, an important question is how strong is the asymmetry between net carbon emissions and removal in an interacting carbon cycle under different removal deployment scales and under different climate and carbon concentration background conditions and how well are these accounted for the model-based assessment of NETs.

Here, we present the results from an anonymous expert survey, including 32 ESM experts and 18 IAM experts, designed to provide insights into (i) how important the two communities consider NETs for ambitious climate policies, (ii) what they consider as the main constraints for the various technologies, (iii) how well the various technologies are represented in their models, and (iv) what kind of (modeling) improvements and exchange is viewed necessary to come up with more realistic estimates for the future climate mitigation contribution of the various NETs.

The article is structured as follows. Section 2 explains the survey. Sections 3 and 4 present and discuss the results, respectively, before the article ends with some concluding remarks.

In an expert survey, we asked ESM and IAM experts about their conceptions of the future role of NETs, including questions related to factors constraining the feasibility of the technologies, and how well the various technologies are represented in current models. The survey ran from 22 November 2018 to 2 January 2019 and was an anonymous online questionnaire (that means we cannot link the responses to individual experts, but we have the information whether invited experts have responded). We invited 115 experts to the survey (USA: 29; Germany: 16; France 10; Japan: 8; UK: 8; Netherlands: 6; Canada: 5; Finland 5 and 28 further from Sweden, Austria, Switzerland, Norway, China, Italy, Greece, Spain, India and Australia). To assemble the invitation list, we identified the most common models (24 ESMs, 25 IAMs) and at least 2 experts per model. The specific IAMs and ESMs covered by in our invitation list can be found in the supplementary information A and the complete questionnaire in supplementary B, both available online at stacks.iop.org/ERL/14/104004/mmedia.

In the questionnaire, we first asked about the type of model respondents mainly use and which specific NETs can be assessed with this model. At this point IAM and ESM experts were filtered into similar but slightly different sets of questions about the representation of carbon removal pathways and carbon cycle feedbacks for five of the most commonly discussed NETs: artificial ocean alkalinization (AOA), afforestation (AF), enhanced weathering on land (EW), BECCS, and direct air capture of CO₂ with storage (DAC). Clearly, we could have included additional NETs, and in particular the prominent role of soil carbon sequestration in currently discussed NETs proposal would have also warranted its inclusion (Minx et al 2018). However, we decided to restrict the questions regarding the removal pathways to five technologies to keep the survey manageable for our time-constrained experts. Furthermore, we aimed for including a broad range of technologies which cover geological, terrestrial, and ocean carbon storage via chemical and biological reactions. As BECCS and AF have already similar pathways, we decided to include EW instead of soil carbon sequestration at this stage.

ESM experts answered the question for all five technologies, IAM experts only for the technologies that their model can assess. We also asked which factors need to be improved to obtain a better representation of the specific NET's removal pathway and (if applicable) the ocean carbon cycle response and the terrestrial carbon cycle response. We provided respondents with a schematic picture with the removal pathways and carbon cycle responses for each method (for the case of BECCS, see figure 1, the corresponding figures for the other NETs can be found supplementary information B).

Zoom In Zoom Out Reset image size

In the final section ESM and IAM experts were asked the same question about the relevance of NETs for climate policy, in general and regarding specific NETs (i.e. which technology would you include in a portfolio). Here, we extended the NETs listed above by soil carbon sequestration (SC), biochar (BC), artificial ocean upwelling (AOUp), ocean iron fertilization (OIF), and blue carbon (BLC) to have a more comprehensive coverage of technologies currently discussed (National Academies of Sciences, Engineering, and Medicine 2018, Royal Society 2018). Asking for all of these technologies about the representation of carbon removal pathways and carbon cycle feedbacks would have simply made the questionnaire too long. Furthermore, we believe that the selected five NETs listed above provide a sufficient representation of carbon cycle pathways and feedbacks affected by carbon dioxide removal. We also asked whether the respondents' models are capable of simulating NETs portfolios and concluded the questionnaire with a set of questions about the constraints for the deployment of specific NETs resulting from non-CO₂ forcing, climate feedbacks, environmental and human health side effects, resource competition, carbon cycle responses and feedbacks, physical CO₂ removal capacity, political feasibility (including acceptance), and cost effectiveness. Throughout the survey, the respondents had the opportunity to obtain information about the specific NETs and could in addition to selecting between options also give free-text answers.

The response rates of IAM experts was rather low compared to ESM experts (27%, i.e. 18 responses from 66 invitations compared 65%, i.e. 32 responses from 49 invitations). Furthermore, four IAM and nine ESM respondents dropped out at some point during the survey. We used the recorded answers of the drop-outs in the analysis as far as possible. Of the 18 IAM-respondents, 16 indicated that they use detailed process IAMs and 2 use benefit-cost IAMs. Of the 18 IAM-respondents, 5 indicated that they rely on impulse-response functions to represent the global carbon cycle in their model, 5 indicated that they rely on nonlinear box-type representations, 3 indicated that they rely on linear box-type representations, 2 indicated that they rely on combinations (of impulse response and (nonlinear) box-type representations, and 3 indicated 'do not know'.

Of the 32 ESM respondents, 22 indicated that they use ESMs of full complexity, 7 use ESMs of intermediate complexity, 2 indicated that they use other models, and 1 indicated 'do not know'. Overall, 21 ESM experts indicated (with multiple answers allowed) that they research one or more components pertaining to the ocean (physics + biogeochemistry + sediments), 15 ESM experts indicate that they research one or more components pertaining to the atmosphere (physics + chemistry), 15 ESM experts indicate that they research the terrestrial biosphere. Table SI.A.T1 in supplementary information A provides further information about the field of research of ESM experts.

3.1. Policy relevance of NETs

The majority of respondents strongly supported the view that negative emissions are essential for ambitious climate policy and they indicated that the contribution of NETs is expected to be achieved rather by a portfolio of technologies than by only one or two technologies (figure 2). IAM experts were slightly less pessimistic about conventional emission abatement as only 5 out of 14 IAM experts consider meeting the 2 °C-goal without NETs as very unlikely, compared to 17 out of 25 ESM experts (figure 2, panel (a)). Panel (b) in figure 2 shows that IAM- and ESM experts had similar conceptions of a future NETs portfolio (i.e. a NETs portfolio as part of the climate policy portfolio with mitigation, adaptation, and negative emissions), but it also shows that IAM experts favored land-based NETs, like AF, BECCS, DAC, and BC above ocean-based NETs. While ESM experts ranked the NETs similarly, they more often included ocean-based methods like BLC and AOA. The most striking differences between the two expert groups was observed for AOA: while only 8% of IAM experts (N = 1) included this technology, 48% of ESM experts (N = 11) indicated that increasing the alkalinity of the ocean should be part of a future NETs portfolio. In contrast, AOUp and OIF were rarely included (only 1 and 3, respectively, respondents included AOUp and OIF). Panel (c) shows that all experts but two actually chose a portfolio (i.e. two or more NETs), and the majority (mode of 8) included six different NETs.

Zoom In Zoom Out Reset image size

3.2. Constraints for NETs

Experts assessed the extent to which eight different factors could constrain the application and feasibility of ten different NETs. Figure 3 shows the mean constraint ratings, differences in mean ratings between the ESM and the IAM experts, and knowledge gaps (via the share of 'do not know' answers). Across all NETs, experts considered removal efficiency (physical removal capacity and cost effectiveness) and political feasibility to be the strongest constraints. On a scale ranging from 0—no constraint, 1—weak constraint, 2—medium constraint, to 3—strong constraint, they indicated the most severe constraints for BECCS (overall mean, 1.75), which was mainly driven by resource competition (2.63), cost effectiveness (2.17), and political feasibility (2.13). In terms of mean constraints OIF and AOUp follow (1.68 and 1.67, respectively), however, both technologies did not receive support to be included in the future NETs portfolio (see figure 2). Disregarding these two technologies, BECCS is, in terms of severity of perceived constraints, followed by two other land-based NETS, EW and AF (with 1.54 and 1.53 mean assessments, respectively). EW was perceived to be most strongly constrained by efficiency, while AF, like BECCS, was perceived to be mainly constrained by resource competition. We found the lowest average constraints for DAC (1.23), SC (1.36), and BLC (1.37). The means for DAC were among the lowest for all constraints except cost effectiveness, which was perceived to be the strongest constraint for DAC (2.62). However, there is a considerable spread among experts assessments: only 2 out of 80 constraint-technology combination did not receive at least once the minimum constraint rating of 0 (AOUp and DAC have a minimum constraint rating of 1 in the constraints carbon cycle responses and feedbacks and cost effectiveness, respectively) and only 3 out 80 contraint-technology combination did not receive at least once the maximum constraint rating of 3 (AOUp, BLC, and OIF have a maximum constraint rating of 2 in the constraint non-CO₂-forcing).

Zoom In Zoom Out Reset image size

In general, ESM experts perceived constraints to be stronger compared to IAM experts: they perceived only 7 out of 80 possible constraint-technology combinations as weaker compared to IAM experts, though these differences are very small in size and not associated with p-values below 0.15 (right panel in figure 3). Ignoring AOUp and OIF, the following three NETs constraints show the largest differences in mean assessments between ESM and IAM experts: resource competition on BLC (${\rm{\Delta }}\bar{x}$ = 1.44; p = 0.002), non-CO₂ forcing on SC (${\rm{\Delta }}\bar{x}$ = 0.94; p = 0.010), and carbon cycle response and feedbacks on BLC (${\rm{\Delta }}\bar{x}$ = 0.86; p = 0.035). This is also partly reflected in the high overall differences for SC (${\rm{\Delta }}\bar{x}$ = 0.53) and BLC (${\rm{\Delta }}\bar{x}$ = 0.54).

The lower panels of figure 3 reflect knowledge gaps: they display the share of 'do not know'-answers. Two patterns emerge: first, the interaction with and effects on non-CO₂-forcing were indicated to be not well known, and second there were large knowledge gaps for ocean-based methods (especially BLC). Many IAM experts did not know about BLC in general and especially about its physical removal potential. Moreover, with respect to BLC and resource competition, their assessment appeared to be the opposite of what ESM experts indicated. Table SI.A.T2 in the supplementary information displays the number of responses for each constraint-technology combination.

3.3. Current capabilities of modeling NETs

Figure 4 summarizes the conceptions of the experts on modeling AOA, AF, EW, BECCS, and DAC in their own model. The experts were asked whether the removal pathways and carbon fluxes are insufficiently, somewhat sufficiently, or even sufficiently represented. However, IAMs usually apply simplified carbon-cycle and climate system representations, approximating various natural fluxes either with box-models or impulse-response functions, whereas ESMs aim at modeling these processes explicitly. Consequently, we applied two filters for the IAM experts. First, they were asked whether their model is capable at all to assess specific NETs and only those who indicated capability were asked about sufficiency. Most IAM experts indicated that their model can assess BECCS, AF, and DAC, only few can assess EW, and one can assess AOA. Second, they were asked whether their model account for the specific removal pathways and carbon fluxes not-at-all, implicitly or explicitly. Only those who indicated implicitly or explicitly were asked about the sufficiency. For the ESM experts we did not apply any filter and asked all ESM experts about the sufficiency of representation of the removal pathways.

Zoom In Zoom Out Reset image size

Figure 4 confirms that ESM experts were much more skeptical about the representation of NETs in their models than the IAM experts. One exception is the atmosphere-to-ocean pathway for AOA. Here, the majority of ESM experts considered the pathways to be at least somewhat sufficiently represented in their model. Clearly, the overall numbers of responses from the IAM experts are low, allowing only an indicative interpretation. For example, there are only few responses for AOA (N = 1) and EW (N = 3) which indicates that the technologies are overall not well represented in IAMs. Furthermore, the models that can assess these technologies either do not include all pathways and responses or do not represent them sufficiently. Most interestingly is the comparison of the two communities for BECCS (which can be assessed by almost all IAMs in our survey).

While the majority of IAM experts indicated that the representation of removal pathways is sufficient, the ESM experts were much more pessimistic. Among them, 7 considered the atmosphere-to-terrestrial pathway insufficiently represented (out of 23) and 13 considered the terrestrial-to-geological pathway insufficiently represented (again out of 23). The stronger confidence in the capabilities of their own model among the IAM experts compared to ESM experts became also apparent in the question about modeling portfolios of NETs. While 70% of the IAM experts indicated that their model can simulate portfolios at least decently, only 50% of the ESM experts conceived their model to be capable of achieving this task at least decently (see figure SI.A.F1 in the supplementary material A).

3.4. Improvements required for modeling NETs

With our invitation list we aimed at a broad representation of currently used models for the assessment of NETs. Given the specificity of the questions, it is possible that experts that are already more familiar with NETs were more likely to respond; this should actually increase the quality of responses. At the same time, the results for our question about which NETs are represented in the IAMs reflects the information from the model descriptions in the supplementary information to chapter 2 in IPCC SR15 (Rogelj et al 2018) very well. Furthermore, table SI.A.T1 indicates for ESM respondents a balanced representation of research foci relevant for the assessment of carbon removal pathways. The fact that fewer IAM experts responded compared to ESM experts might be an indication that the IAM community is still less familiar with NETs.

Our results are a glimpse into the conceptions of the experts who actually responded and their models. The results should thus not be extrapolated to all modelers and all models but rather to those who actually model NETs. We therefore only use statistical tests at one instance where we compare ESM and IAM experts. The rather low number of responses obtained from the IAM community limits the possibilities for statistical interference; this means that low p-values derived in section 3 should not lead readers to ignore the remaining uncertainty about the 'true' beliefs and conceptions in the IAM community (and similarly in the ESM community); but also means that higher p-values should not lead readers to overlook interesting differences in the conception of the two communities (Wasserstein et al 2019). Overall, we are confident that our results are helpful for the interpretation of model-based NETs assessments for climate policy and for future model and scenario development with respect to NETs.

This holds in particular true given the differences in the conceptions about NETs between the two expert groups. Two of the most striking differences occur for the assessment of AOA and BLC: while only 1 IAM expert included AOA, about half of the ESM experts indicated that increasing the alkalinity of the ocean should be part of a future NETs portfolio. This does, however, not speak to a general reservation against ocean-based NETs, as 38% of IAM experts would include BLC in a NETs portfolio (compared to 52% of ESM experts). Looking more closely at the assessment of factors that could potentially constrain the use of the NETs, we furthermore found that IAM experts perceived the physical removal capacity a much smaller constraint for BLC deployment compared to ESM experts (1.63 compared to 2.2 with 3 indicating the maximum constraint). In the assessment reports by the Royal Society (2018) and the National Academies of Sciences, Engineering, and Medicine (2018) AOA has an estimated annual removal potential of 40 Gt CO₂ compared to 0.13 Gt CO₂ for the case of BLC. The large physical potential of AOA does not necessarily translate into a large economic potential and in turn the rather low share attributed to this technology by IAM experts might be based on such kind of economic reasoning. Still, taking into account that IAM-based assessments of NETs predominantly focus on reforestation, afforestation and BECCS, it appears that ocean NETs have so far mostly been overlooked by the IAM community.

The current focus on BECCS appears somewhat surprising because our experts indicated that this technology faces the strongest constraints in terms of feasibility. However, it might just because BECCS has already been researched and discussed in more detail (including for example also public acceptance research); and thus problems and constraints are simply better understood. Accordingly, other NETs with a lower technology readiness level might have received less attention and consequently less scrutiny; this could mean that their assessment is still too optimistic.

However, the pessimistic assessments of BECCs could also indicate that intrinsic conflicts regarding land and water competition are considered to be almost insurmountable, significantly limiting the application of the technology in the future. This becomes also apparent when comparing BECCS with DAC—the latter being assessed by the experts to be much less constraint in terms of resource competition, suggesting that the carbon capture and (geological) storage aspect of BECCS is not necessarily considered to be a limiting factor for BECCS. Furthermore, comparing BECCS with BC and SC (the latter being among the most research NETs since 1990, Minx et al 2018), it becomes apparent that land and water conflicts are considered to be limiting BECCS because BC and SC were not only considered to be less constraint overall but also in terms of resource competition.

Only 62% of IAM experts included SC in their future NETs portfolio, compared to 92% who included BECCS. ESM experts included SC just as often as BECCS in their NETs portfolio (74% and 70%, respectively, see figure 2). Experts seem to be aware that annual carbon removal rates at the potential scale of BECCS cannot be achieved via SC indefinitely because of soil saturation (Royal Society 2018): they assessed the physical removal capacity for SC as its most limiting constraint. SC was assessed to face also the highest constraint on removal capacity compared to the other terrestrial and (partly) biological-based NETs.

Among all constraints, NETs were assessed to be strongest constrainted by cost effectiveness. Excluding DAC from the list of NETs, cost effectiveness remains the strongest constraint, closely followed by political feasibility (averaging to 2.07 and 1.89 versus 2.01 and 1.95 with and without DAC, respectively). Even with cost effectiveness included as constraint, DAC was assessed by experts to face the lowest constraints. Consequently, the actual contribution in terms of carbon removal by DAC in future NETs portfolio could be significant if recent cost estimate updates, suggesting that the cost could drop below 100USD/tCO₂ (Keith et al 2018) (compared to previous estimates of about 600 USD/tCO₂) prove to be achievable in practice.

In general, ESM experts perceived constraints more pessimistically when compared to IAM experts: they perceived only 7 out of 80 possible constraint-technology combinations as weaker compared to the IAM experts (see right panel in figure 3). These (and other differences) might be driven by diverging conceptions of the technologies or different levels of knowledge in the two communities. ESM experts might misjudge socio-economic factors. For example, it appears surprising that ESM experts considered the political feasibility of BC and EW to be a much stronger constraint than the IAM experts. On the other hand, IAM experts might lack knowledge about biogeochemical processes, e.g. of likely side effects, which could lead them to underestimate certain constraints as well.

A similar pattern can be observed regarding the question how well the expert's own model captures aspects of the global carbon cycle relevant for NETs. While the majority of IAM experts indicated that the representation of removal pathways is sufficient, the ESM experts were much more pessimistic. Among them, 7 considered the atmosphere-to-terrestrial pathway insufficiently represented (out of 23) and 13 considered the terrestrial-to-geological pathway insufficiently represented (again out of 23). Again, this might be explained by different modeling paradigms in the two communities. While ESM experts often aim to explicitly model the various underlying physical, chemical, and biological processes, IAM experts are rather interested in properly measuring the net contribution to climate change mitigation than the exact representation of a pathway. Furthermore, ESM experts spend a lot of time validating their models against historical observations, providing them with more knowledge about how 'bad' their models are. Consequently, the representation of carbon removal pathways in IAMs might not necessarily be worse compared to the representation in ESMs. However, they are probably still insufficient to reliably answer the question about the optimal future shares of specific NETs because the ESMs that are used to calibrate simple climate-carbon models in IAMs also still fail to properly answer the question.

Our results emphasize that we need to understand the limitations and implications of a broad set of NETs and their interactions in technology portfolios as soon as possible, to conceive realistic climate mitigation pathways. Current ESM- and IAM-based investigations still focus on rather stylized scenarios or are restricted to afforestation, reforestation, and BECCS, respectively. A reason for the latter may be that planting trees is perceived as a benign method and hardly met with public resentment (Braun et al 2018) even by those who are against the use of NETs. Yet, this approach ignores a large number of potential NETs under discussion, and limits the study of portfolios and interactions between different NETs. Current IAM-based climate policy advice appears to potentially overestimate the future negative emissions potentially achievable via BECCS, in particular because experts themselves consider resource competition, cost effectiveness, and political feasibility to be strong constraints for future BECCS deployment. IAMs rarely investigate NETs portfolios, implying that the future negative emission contribution via NETs portfolios is potentially underestimated, in particular because IAM-based assessments neglect the contribution of ocean NETs. Also many carbon removal simulations in ESM are rather idealized, missing some important pathways and components and allowing answering only basic questions. Specific NETS like enhanced weathering or biochar cannot be adequately simulated in ESMs. In the process of improving both, ESMs and in turn IAMs, stronger exchange between the ESM and IAM community would be beneficial for a more comprehensive assessment of NETs. Ideally such exchange is part of joint research programs on model development and comparison.

We would like to thank all experts who volunteered their time and participated in the survey. We would like to thank the editors of Environmental Research Letters and two anonymous referees for helpful suggestions and comments. Wilfried Rickels acknowledges funding received from the German Research Foundation (project CDRecon, 390288545, RI1833/4-1) and from the Research Council of Norway (project IMPOSE, Grant Nr. NFR-294930). Fabian Reith acknowledges funding received from the German Research Foundation Excellence Initiative (Cluster of Excellence 'The Future Ocean', project CP1689 'Ocean Pollution'). David P Keller acknowledges funding received from the German Research Foundation's Priority Program 1689 'Climate Engineering' (project CDR-MIA; KE 2149/2-1). Christine Merk acknowledges funding received from the German Research Foundation's Priority Program 1689 'Climate Engineering' (project TOMACE; 311117145). The usual caveats apply.

The data that support the findings of this study are available from the corresponding author upon reasonable request. The data are not publicly available for ethical reasons.