A deep dive into the modelling assumptions for biomass with carbon capture and storage (BECCS): a transparency exercise

Bioenergy with carbon capture and storage (BECCS) is envisaged as a critical element of most deep decarbonisation pathways compatible with the Paris Agreement. Such a transformational upscaling—to 3–7 Gt CO2/yr by 2050—requires an unprecedented technological, economic, socio-cultural and political effort, along with, crucially, transparent communication between all stakeholders. Integrated Assessment Models (IAMs) that underpin the 1.5 °C scenarios assessed by IPCC have played a critical role in building and assessing deep decarbonisation narratives. However, their high-level aggregation and their complexity can cause them to be perceived as non-transparent by stakeholders outside of the IAM community. This paper bridges this gap by offering a comprehensive assessment of BECCS assumptions as used in IAMs so as to open them to a wider audience. We focus on key assumptions that underpin five aspects of BECCS: biomass availability, BECCS technologies, CO2 transport and storage infrastructure, BECCS costs, and wider system conditions which favour the deployment of BECCS. Through a structured review, we find that all IAMs communicate wider system assumptions and major cost assumptions transparently. This quality however fades as we dig deeper into modelling details. This is particularly true for sets of technological elements such as CO2 transport and storage infrastructure, for which we found the least transparent assumptions. We also found that IAMs are less transparent on the completeness of their treatment of the five BECCS aspects we investigated, and not transparent regarding the inclusion and treatment of socio-cultural and institutional-regulatory dimensions of feasibility which are key BECCS elements as suggested by the IPCC. We conclude with a practical discussion around ways of increasing IAM transparency as a bridge between this community and stakeholders from other disciplines, policy decision makers, financiers, and the public.


Introduction
Integrated Assessment Models (IAMs) are complex frameworks bringing together knowledge from several disciplines, e.g. energy systems modelling, land use, macroeconomics, and climate modelling (IPCC 2014). Their broad scope has made them very useful tools for designing and analysing scenarios of future global decarbonisation pathways, and IAMs have played a critical role in underpinning long-term climate change mitigation assessments (IPCC 2014) commissioned by the Intergovernmental Panel on Climate Change (IPCC). This has brought IAMs high scientific visibility (IPCC 2018), but also put them under intense scientific scrutiny, especially related to the transparency of their data and modelling assumptions (Pindyck 2017, Weyant 2017, Gambhir et al 2019. A focal point of this scrutiny has been on the models' reliance on biomass with carbon capture and storage (BECCS) to meet deep decarbonisation pathways especially in the latter half of the 21st century. Indeed, BECCS is the critical element of the majority of 2°C or 1.5°C compatible pathways (IPCC 2013(IPCC , 2018. It is also simultaneously the most multi-disciplinary (Smith et al 2016) and most controversial technology (Fuss et al 2014). IAM results that include large scale deployment of BECCS have been scrutinised from an inter-generational equity perspective, i.e. near-versus long-term climate mitigation (Anderson and Peters 2016, Obersteiner et al 2018, adverse impacts on other resources (Smith et al 2016), land use competition and social acceptability (Vaughan and Gough 2016), ethical issues and risk of use (Lawrence et al 2018), and the sheer scope of both innovation and upscaling required from an immature technology (Lenzi et al 2018, Nemet et al 2018. Notwithstanding, there was recognition that there is only a partial coordination between IAM modellers and other disciplinary experts who operate at a more detailed level of aggregation (Minx et al 2017).
To help bridge this gap between IAM modellers and broader disciplinary experts, our study examines the transparency of assumptions for the deployment of BECCS in IAMs. We conduct a structured review of six of the leading IAMs, one of which is our in-house IAM (TIAM-UCL), for which we have complete access to the underlying assumptions and documentation. To maintain an objective view on the transparency of assumptions in IAMs, including TIAM-UCL, we adopted a neutral position, in the sense that we reviewed assumptions that were publicly available, but we did not contact individual IAM modelling teams. This allowed us to test what non-modellers can actually see when they try to achieve a deeper understanding of IAM results and of the assumptions that underpin them. The aim of this transparency exercise is to offer guidance on model transparency to support the interpretation and comparison of future results. This should both enable an improved dialogue between the IAM community and different research communities (Geels et al 2016). It should also improve the integration of quantitative and qualitative insights  for example along the (complex) supply chain of BECCS. This paper is structured as follows: section 2 describes the methods we employed to undertake this review. Section 3 contextualises the most transparent key BECCS assumptions in TIAM-UCL as compared to other IAMs and sets the scene for the deeper transparency analysis that follows. Section 4 uses a traffic light categorisation to examine the transparency of underlying constraints and drivers of BECCS. Full explanatory details are found in the appendix. Section 5 widens the discussion and highlights what is not included in the scope of the model (but instead is implicit) and (from an alternate disciplinary viewpoint) may be very important. Section 6 summarises findings of both transparency and the critical examination of key assumptions around BECCS, concluding with recommendations for increasing model transparency.

Methods for reviewing the transparency of BECCS assumptions in IAMs
Given their complexity, dissecting the highly detailed model structures and assumptions of IAMs is not straightforward. This is a well-known analytical problem, which requires up-to-date transparency (DeCarolis et al 2012) rather than a reliance on knowledge on past model versions and sources (Dodds et al 2015).
The complexity of BECCS adds a further challenge to investigating modelling assumption transparency. Firstly, BECCS is not an industrial technology with established efficiency. Instead, the term covers an entire supply chain, from cultivating and harvesting biomass to producing different biofuels. It also covers CO 2 capture, liquefaction, as well as its transport to, and injection into geological storage. Modelling assumptions need to be made at each stage of this supply chain, all of which are sector-, space-and timespecific.
Secondly, upscaling BECCS from its current level of 1 MtCO 2 /yr to those envisaged by IAM scenarios will require the fast ramping up of activities across the full supply chain. This assumes that all the markets involved whether for 'biomass for energy', biofuel commodities, or CO 2 function smoothly at both national and global levels (Lenzi et al 2018). Modelling assumptions on growth are usually sector, time-, and location-specific. Each of these will also be adjusted depending on views of future policy and socio-economic pathways.
Finally, in addition to providing low carbon fuels, BECCS is also assumed to provide 'negative emissions'. This means that the overall balance of GHG emissions over the full supply chain of BECCS is assumed to be negative. Understanding the transparency of this assumption relies on being able to assess the underlying assumptions that describe the full carbon balance of each individual step. This means reviewing the uptake of CO 2 by biomass growth; the GHG emissions from biomass cultivation, harvest, storage and processing; the efficiency of processing; the energy required for capturing, transporting and storing CO 2 as well as the carbon losses along the way.
We started the transparency review by comparing well-communicated BECCS assumptions in IAMs as reviewed by Fuss et al (2018) versus TIAM-UCL assumptions. These include BECCS costs, and the magnitude of global biomass production and CO 2 storage potentials. This comparison summarises the range of potentials and costs across the IAMs so as to guide further investigation of underlying constraints and assumptions. In a second step, we take advantage of our combined expertise in, and full knowledge of, TIAM-UCL to structure our review for specific parameters along the BECCS supply chain (tables 5-9 is available online at stacks.iop.org/ERL/15/084008/ mmedia in section 4). As much as possible, these were selected to cover the complexity of BECCS, including carbon accounting over the full BECCS supply chains. The transparency of assumptions is characterised using a traffic light system. Green lights represent BECCS aspects that are well communicated by the modellers (including ourselves), amber ones denote partial communication or transparency, and red characterises those aspects that are or not transparent or not communicated. Transparent communication of parameter assumptions however implies that this respective parameter is included in the modelling framework that is under scrutiny. Accounting for the fact that some parameters are not included across all modelling frameworks, the traffic light system was adjusted so that: green lights represent BECCS assumptions which are included in the modelling framework and are well communicated by the IAM teams; amber ones denote that the parameter is included, but there is no clear communication of assumptions (partially specified assumption, or conflicting information coming from different sources, e.g. web documentation referring to several external documents); red means that the parameter is not specified at all and is potentially not included in the modelling framework. The basis for this quantification is what is written in the model documentation and key journal papers. It does not rely on any understanding of the full historical evolution of the structure of the models or of their application (Dodds et al 2015). A full and detailed discussion of the transparency assessment (green, amber, red) is given in the appendix. To summarise the transparency findings, we assign each colour a score, i.e. green is assigned 1, amber 0.5 and red 0. A transparency score is then calculated for each IAM in each of the five BECCS aspects investigated here by dividing the sum of all its colour scores by the maximum score which could be obtained for that aspect, i.e. if all the parameters were communicated transparently.
We chose six leading IAMs: IMAGE, MESSAGE-GLOBIOM, GCAM, REMIND/MAgPIE, AIM, and TIAM-UCL. For each IAM, we have considered the model documentation and recent journal publications relevant to the deployment of BECCS under global deep decarbonisation scenarios (1.5°C and 2°C), see table 1. We also considered model inter-comparison studies published by the IAM teams, the SSP database hosted by IIASA, and the recently released IPCC SR1.5C database. Our main criteria in examining each model's documentation and selected studies was that they should provide enough transparent information for a well-versed reader to scrutinise their BECCS assumptions. If a parameter or a parameter value is not easy to find, it means the information is not transparently communicated.
We explicitly acknowledge that the number of studies we reviewed is limited due to practical reasons, but it is fit for purpose. It shows how easy, or complex, is to find key assumptions when you are a third-party, not directly involved in the development and running of IAMs, but wishing to contribute to the BECCS debate.

Key IAM assumptions on BECCS
Key BECCS assumptions which are usually well communicated in IAM studies include BECCS costs and the global magnitude of both biomass resource and CO 2 storage (Fuss et al 2018). Based on these, each IAM estimates the global BECCS potential under different futures (shared socioeconomic pathways (SSPs)) and different projections of global GHG emission concentrations (representative concentration pathways (RCPs)). This section compares these aggregated assumptions (see table 2) to those made in the database and code of our in-house IAM (TIAM-UCL) to which we have full access. This then leads us to an in-depth examination of the underlying model constraints and drivers of these assumptions in the six selected IAMs (in section 4), which is our main contribution.

Global biomass potential
The global biomass potential is reported as being a key limiting factor for the large scale deployment of BECCS , IPCC 2014, IPCC 2018. The global biomass resource base in IAMs is usually composed of several biomass fractions, e.g. dedicated energy crops, agricultural and forest residues, and waste fractions. There is high uncertainty surrounding the availability, economic feasibility and sustainability of all these fractions (Fuss et al 2018), but the largest and most debated fraction, are energy crops. These usually include herbaceous and woody crops cultivated purposely for energy use. The global potential for energy crops is driven by agricultural development (i.e. yield increase) and by the availability of land for bioenergy production. The latter is subject to constraints that relate to (i) competition for land with other human uses, e.g. food, timber, conservation purposes; (ii) ecological limits, such as water scarcity, soil degradation or biodiversity protection; and (iii) issues of biomass seasonality and storage. Modelling assumptions made around each of these constraints combine to produce a wide range of possible biomass potentials. We investigate the transparency of these underlying assumptions in section 4.

Global CO 2 storage potential
Following the Global CCS Institute (2016), there is enough global storage available for CO 2 captured from Afforestation option CCS representation CO 2 capture, transport and storage modelled individually. Regional differentiation of CO 2 transport and storage costs.
No regional differentiation of CO 2 transport and storage costs. One global geological reservoir.
Regional differentiation of CO 2 transport and storage costs.
Fixed CO 2 transport cost. Region and storage specific CO 2 storage costs.
Fixed carbon capture costs. CO 2 transport and storage costs not specified.
Fixed CO 2 transport cost. Regional differentiation of storage capacity and costs.
Selected publications van Vuuren et al (2011van Vuuren et al ( , 2013 biofuel and fossil sources, especially when including offshore potentials. However, as indicated in table 3, there is a large uncertainty around where this storage will be made available and the potential mismatch between production of CO 2 and available storage sites leading to a global cumulative storage potential of 2100 Gt CO 2 . The main difference in geological storage assumptions relates to potentials available in aquifers for which TIAM-UCL does not include trapping mechanisms. Note that, independently of its potential, the actual use of CO 2 storage may also be subject to other factors such as: the development of a CO 2 transport infrastructure, the public acceptance of CCS, the total cost of preparing the storage site, or that of monitoring and verifying the permanence of the storage (Haszeldine et al 2018). These topics are further investigated in the next section.

Costs of BECCS
Based on a systematic review of the literature and on expert judgement, Fuss et al (2018) estimates the cost of BECCS in 2050 to be in the range of 100-200 $/t CO 2 . These estimations account for how difficult it is to access biomass, for the cost of land and its conversion, for the type of bioenergy facility, and for the CCS infrastructure required, see table 4. TIAM-UCL estimates for these costs all fall in the same range with the exception of using BECCS for the production of advanced (Fischer Tropsch) biofuels. These are 50% higher, mainly due to the cost of the biomass and to both technology type and efficiency. These cost assumptions influence the affect the extent to which BECCS is used in decarbonisation scenarios, i.e. how many tonnes of CO 2 BECCS technologies remove per year in these alternate futures. The aggregated assumptions discussed above are usually published in papers and reports from the IAM community. Based on knowledge of TIAM-UCL, the next sections proceed to unravel the underlying constraints and drivers that underpin these assumptions but that are not usually disclosed or discussed.

Deeper assessment: underlying constraints and drivers of BECCS
In this section, we focus on the transparency of underlying constraints and drivers that relate to assumptions under scrutiny. A traffic light system is used for visual clarity. Green denotes BECCS aspects which are well communicated by IAM teams, amber denotes partial communication, and red denotes that these are not communicated with model results. The data values and modelling assumptions presented in each table are described in the appendix together with our comments on the transparency of communication. We follow the full supply chain of BECCS, starting with biomass potential (table 5), bio-technologies with carbon capture (including biomass to energy transformation and capture of CO 2 , table 6), CO 2 transport and storage (table 7), and costs across

Model
Global potential (Gt CO 2 ) Regional potential (Gt CO 2 ) Depleted oil and gas fields the BECCS supply chain (table 8). We investigate the transparency of carbon accounting in IAM modelling by including the GHG emissions that correspond to successive steps in the BECCS supply chain in each of the tables. We also include a table compiling crosscutting issues that influence the use of BECCS for climate mitigation (table 9).

Biomass potential
Future global biomass potential is highly uncertain because it depends on techno-economic, environmental and social factors which are complex as well as region and time dependant. In this section we investigate assumptions around land competition, yields of energy crops, ecological constraints, and biotrade which determine the magnitude of the biomass that is available for energy. We also dig into the details of carbon accounting. Shown in table 5, our results reveal that all the IAMs we assess are transparent around land competition and energy crops productivity. Different to TIAM-UCL, which has a simplified exogenous model of land use, all the IAMs we review include a spatially explicit representation of the competition for land between food, energy and forestry. The modelling teams share the resulting land allocation for energy crops transparently both in model inter-comparison studies, e.g. Popp et al (2017), and model specific publications, e.g. Doelman et al (2018). It is interesting to note that under a SSP2-2.6 scenario (a 'middle-of-the-road' future with a climate forcing of 2.6 W m −2 in 2100), the land allocated to biomass for energy ranges from 225 Mha in IMAGE to 1100 Mha in GCAMv4 (table 5(a) in the appendix). This is due to a combination of low (IMAGE) versus high (MESSAGE) sensitivity of food demand to food prices (Popp et al 2017), and to the inclusion of sustainability criteria in IMAGE which limit the expansion of energy crops to lands that are not used for food production. In terms of yield assumptions, all IAMs, except TIAM-UCL, estimate energy crop yields endogenously. TIAM-UCL starts with 2015 regional yields as reported in Ricardo-AEA (Ricardo-AEA 2017) and then assumes 1.3% yield increase per year. This leads to regional yield values of between 5 and 12 dry tonnes/ha by 2100 as compared to 11 dry tonnes/ha estimated by IMAGE and GCAM, 14 in MESSAGE-GLOBIOM, and 21 in both AIM-PLUM and REMIND-MAgPIE. Looking at ecological constraints, i.e. water scarcity, soil and biodiversity concerns, we found that four out of six IAMs explicitly account for them (green in table 5), while the others, including TIAM-UCL are vaguer on this topic (yellow in table 5, see appendix for more details). While these ecological constraints could reduce both yields and land suitability for energy crop production, we found no explicit quantification nor any communication of how much they could affect regional and aggregate biomass potentials. Collaboration and trade between the different regions is essential to BECCS deployment, especially under stringent climate scenarios. Looking at how transparently the trade assumptions are communicated by IAM teams, we found that the type of biomass and biofuels for trade is fairly visible in all IAMs. However, the assumptions on trade links between regions and how they evolve under alternate future scenarios are less visible or not communicated by several IAMs.
Assumptions around carbon accounting in the biomass production stage are one of the main determinants of the potential carbon sequestration by BECCS. van Vuuren et al (2013) report that considering an emission factor of 15 kg CO 2 /GJ produced biomass reduces BECCS effectiveness by a fifth. Our results in table 5 show that while land use and land use change emissions are well represented in all IAMs, biomass storage and transport emission assumptions are either not included or vague. For example, while domestic transport of biomass is spatially explicit in GLOBIOM (Valin et al 2015), the corresponding transport emissions are not specified.

Bioenergy with carbon capture technologies
All the IAMs we reviewed include BECCS for the production of power, bio-liquids and hydrogen. Independently of the type of BECCS available in each IAM, all models usually make assumptions regarding the earliest implementation of these technologies, their build rate (how fast new capacity can be added each year), their availability factor (fraction of time the plant is operating), efficiency of transformation and how this evolves over time, and CO 2 capture rates. These technical assumptions (table 6) are not as visible as land assumptions (table 5). For example, only IMAGE reports a 36 month construction time for biopower generation with CCS (Black&Veatch 2012, LAZARD 2015). All the global IAMs assume that the conversion efficiency of technologies increase over time, albeit with significant variations in the magnitude of the increase (Krey et al 2019). Note that these efficiencies are usually exogenous inputs to the models based on average values taken over different technologies in operation, i.e. not theoretical efficiencies . REMIND-MAgPIE and GCAM are transparent on their assumptions regarding plant life, capacity factor, efficiency of transformation and CO 2 capture rates. It is interesting to note that GCAM has been transparent regarding updates of BECCS technologies e.g. they reduced the efficiency of BECCS for power from 41.6% (Luckow et al 2010) to 18% for a biomass steam plant+CCS, and to 25% for a biomass IGCC+CCS (Muratori et al 2017a). Generally, the technological updates in GCAM have reduced the technological potential of BECCS (see table 6, and in the appendix), but these updated values are still slightly more optimistic than in REMIND-MAgPIE (Luderer et al 2015), and TIAM-UCL.
It is interesting to note that the all the IAMs we assessed assume that bioenergy is carbon neutral, i.e. that the CO 2 emissions linked to producing and using bioenergy in any form are equal to the CO 2 that is sequestered by growing the biomass. Whilst there seems to be general agreement that sustainable biomass growth does re-capture the CO 2 that results from the combustion of biomass, the sequestration and emission rates might be in temporal imbalance (Lamers and Juninger 2013, EASAC 2019, Torvanger 2019. For woody biomass, scientific evidence shows that the time lag between biomass harvest and biomass growth to pre-harvest as compared to not harvesting the biomass (usually termed 'carbon parity time') could be anywhere between 0 and hundreds of years, depending on the biomass resource and on what the resulting bioenergy substitutes (Lamers and Juninger 2013).

CO 2 storage, including transport of CO 2 to storage
Usually IAMs report regional CO 2 storage capacity, sometimes per type of geological storage (table 7, and in the appendix). Note that the geological storage of CO 2 is shared between BECCS, fossil CCS, and other negative emission technologies if available, e.g. Direct Air Capture. The injection rates of CO 2 captured from BECCS are usually communicated, although mostly at global level, e.g. 0-10 GtCO 2 /y in 2050 and 10-20 GtCO 2 /y in 2100 . The wider policy audience would benefit from more transparent assumptions around the preparation and use of geological storage. We have not found any reporting of CO 2 leakage rates, nor monitoring, reporting and verification (MRV) mechanisms to ensure that the stored CO 2 is kept in the geological storage.
The biggest gap in reporting transparency of BECCS modelling assumptions concerns the CCS infrastructure, which connects the CO 2 capturing plants to the geological storage. Except for MESSAGE-GLOBIOM which reports the assumed length of CO 2 pipelines (Riahi et al 2007), we have not found any mention of assumed availability, efficiency, or build rate of CCS pipeline networks in different regions. It seems that most models (including TIAM-UCL) are modelling the CCS infrastructure based on costs estimated by Hendricks et al (Hendriks et al 2004), subsequently updated with other reports, e.g. from EMF28 (Weyant et al 2013). These updates are however not usually made clear. Instead, all IAMs take a rather binary view of CCS availability, running sensitivity analyses assuming, for example, the absence of BECCS in the system because of challenges in developing the CCS infrastructure (e.g. (Bauer et al 2018)).

BECCS costs
All IAMs investigated here have endogenous estimations of the costs of primary biomass for energy, e.g. considering yields, regional land prices and regional income (IMAGE, (van Vuuren et al 2009)), or as a function of capital, labour and intermediate costs (AIM (Hasegawa et al 2017)). In TIAM-UCL the cost of primary bioenergy is given using supply-cost curves derived from Ricardo-AEA (Ricardo-AEA 2017). When we dig into the detail of land rental rates per region and agricultural subsidies assumed for bioenergy production, the transparency of model assumptions decreases, with only some models detailing these costs, e.g. REMIND-MAgPIE applies a bioenergy tax, rising from 0% in 2030 to 100% in 2100, to reflect sustainability concern, while IMAGE adds explicit energy taxes and subsidies at both the primary and end-use level (PBL 2014). We also found (table 8) that assumptions on the cost of storing and processing biomass prior to its transformation into energy are usually not available: GCAM is the only IAM to report average biomass processing costs of $1.87/GJ, or $36.5/tonne biomass (Luckow et al 2010), while none of the IAMs report biomass storage costs (including TIAM-UCL). International transport costs usually result from endogenous model calculations, but it is not clear whether they reflect only the cost of fuels used for transport or if they also account for temporary storage and handling in the ports. Domestic transportation is less accurately represented, except for MES-SAGE-GLOBIOM, which calculates it endogenously based on distance and mode of transport (Valin et al 2015). IMAGE and GCAM consider fixed transport costs per GJ biomass, US$ 0.5/GJ (IMAGE (van Vuuren et al 2011)) versus $0.37/GJ, or $6/tonne biomass in GCAM (Luckow et al 2010).
IAMs make more transparent assumptions about the capital, fixed and variable costs of operating bioenergy technologies with carbon capture. These assumptions are visible in individual IAM publications and are also specified in the inter-model comparison . Similar to (Krey et al 2019) we found that the variation of capital costs is quite large between IAMs, and that the O&M costs are usually given as a percentage from the CAPEX, which is constant both across the regions and in time. In IMAGE, the web documentation points to several data sources which, in turn, lead to a range of different data assumptions. For example, the sources for the CAPEX of BECCS for power are:  (IRENA 2015). It is interesting to note that GCAM differentiates between the costs of capturing high versus low purity CO 2 : 72 2010 $/tCO 2 for a biomass steam plant+CCS, 66 2010 $/tCO 2 for biomass IGCC+CCS, 32-70 2010 $/tCO 2 for cellulosic ethanol with CCS, and 32-46 2010 $/tCO 2 for FT biofuels +CCS (Muratori et al 2017a). Also AIM-PLUM assume 100-150 $/tCO 2 for the manufacturing sector and 50-120 $/tCO 2 for the power sector (based on IEA 2008). Ultimately, the IMAGE web documentation suggests 35-45 2005 $/tCO 2 captured. The other IAMs report CAPEX costs that include the capture of CO 2 . This increases the cost of energy production by about 50% (Hendriks et al 2004).
While the technical assumptions on CO 2 transport and storage are less transparent (table 7), the cost assumptions of these stages are both very visible (table 8) and quite similar between models. For example IMAGE assumes region and storage specific CO 2 transport costs of between 1 and 30 2005 $/tCO 2 , with the majority remaining below 10 2005 $/tCO 2 (Hendriks et al 2004). TIAM-UCL uses similar values, between 1 and 10 2005 $/tCO 2 . MESSAGE-GLOBIOM reports 7-9 2005 $/tCO 2 for fossil CO 2 and double values for biogenic CO 2 , as BECCS plants are smaller than their fossil counterpart, requiring more infrastructure to transport CO 2 to storage (Koelbl et al 2014). REMIND-MAgPIE suggests 3.1-4.2 million $/km CO 2 pipeline (Bauer 2005), which translates to 8-15 $/tCO 2 metric, considering an average pipeline length of 1000 km, 10-15 mtCO 2 transported per year (Bauer 2005) and pipeline operation lifetime of between 20 and 25 years.

Cross-cutting issues
Several cross-cutting assumptions in IAMs, such as the availability of other Carbon Dioxide Removal technologies or the date of peak emissions, will influence the use of BECCS for climate mitigation, see table 9.We found that all IAMs do very well at communicating the stringency of climate targets, i.e. the date at which the system reaches net zero CO 2 emissions, which is usually after 2070. They also communicate transparently that corresponding trajectories of global CO 2 emissions would peak in 2020. All the IAMs recognise that climate mitigation is biased towards supply side measures, e.g. increased efficiency, fossil fuel substitution by renewable fuels, or the use of negative emission technologies (NETs). The NETs usually included in IAMs are afforestation/reforestation and BECCS. The carbon prices are usually uniform across all regions, but the application of regional GHG emission caps can also lead to regional carbon prices, e.g. in MESSAGE GLOBIOM (Fricko et al 2017). It is interesting to note that the general discount rate applied in IAMs is 5%, versus 3.5% usually considered in TIAM-UCL. Finally, IMAGE is the only IAM that mentions the inclusion of the disruptive impacts of climate change on the system through e.g. extreme weather events.

Broader assessment: what is not included/or missing from IAMs
Thinking about the feasibility of different mitigation options, the IPCC suggests a framework for their full assessment across six dimensions: (i) geophysical; (ii) environmental-ecological; (iii) technological; (iv) economic; (v) socio-cultural; and (vi) institutional (IPCC 2018).
On the feasibility of BECCS (table 4.11, IPCC (2018)), the report notes that geophysical and technological dimensions have neither a negative nor a positive effect. Conversely, it highlights potential feasibility barriers in the remaining four of the six dimensions including: environmental (biomass availability), economic status, legal framework for operating BECCS, and social acceptance.
Following our results in table 5, the IAM teams are largely transparent in communicating assumptions in the geophysical dimension. IMAGE probably has the most comprehensive coverage, including terrestrial and aquatic biodiversity, flood risks, land degradation, and ecosystem services (Doelman et al 2018). REMIND MAgPIE has pushed the boundaries of geophysical domain representation in IAMs by assessing the deployment of BECCS within the nine planetary boundaries (Heck et al 2018). In particular, they include biosphere integrity, biogeochemical flows, and fresh-water use required for large scale biomass plantations. MESSAGE-GLOBIOM also includes soil quality and water scarcity and their potential impact on biomass production. AIM explicitly includes biodiversity and soil protection when assessing the global bioenergy potential (Wu et al 2019). However, the geophysical dimension is less transparent in the other IAMs, but implicitly assumed favourable, e.g. GCAM assumes that under a SSP2-RCP2.6 future 1100 Mha of land are suitable for biomass production, five times more than in IMAGE (Popp et al 2017).
Following our results in table 6, IAMs are more opaque in their technological assumptions on bioenergy with carbon capture, while assumptions on the CO 2 transport and storage infrastructure development and roll out are rather absent from all the IAMs we reviewed here (table 7). BECCS are still in their infancy and there are largely unknown risks associated with their large scale deployment (Obersteiner et al 2018). We have not found any IAM communication of technology readiness level and scalability of different types of BECCS in different regions, assumptions which seem critical for a large scale roll out of BECCS (IPCC 2018).
Biomass availability is determined by the competition for land between food, energy, and other human uses, including ecosystem restoration. Considering our results in table 5, IAMs are transparent in communicating assumptions around the competition for land. Future developments of land use are heavily influenced by parameters such as crop yields and livestock intensification (Popp et al 2017). Intensification of land use (or land sparing), as well as its opposite, agricultural expansion, are driven by complex factors such as institutional, government, regulatory and market based instruments, type of land, income of stakeholders, etc (IPCC 2014). These factors cut across the six dimensions indicated by the IPCC and are region and context specific. They are not usually represented in IAM frameworks, but are implicitly assumed to be in place.
One of the most critical aspects of BECCS is their ability to deliver 'negative emissions' on the timescales envisaged by the IAM scenarios, i.e. to 2100 and beyond. If managed sustainably, bioenergy could contribute to global decarbonisation in the long-term, i.e. after 2050 (IEA 2017). This assumes the CO 2 emissions caused by biomass harvest would be sequestered over the life-time growth of biomass. This could be the case when harvesting fast growing woody plantations on unused or degraded land, or harvesting processing residues and standing deadwood from insect infested sites. However, harvesting currently unmanaged forests or replacing forests by fast growing plantations could result in carbon debts which could not be 'paid back' this century (Lamers and Juninger 2013). Furthermore, the efficiency of bioenergy for climate change mitigation is conditioned by what it substitutes at the point in time when it becomes 'carbon neutral'. With the fast increase of cheaper renewable energy options, betting on bioenergy on the long term might result in more emission rather than sequestration. In any case, informed decision making should always consider regional forest carbon balances (Lamers and Juninger 2013) and wider system impacts and counterfactuals of the whole forest and its products (EASAC 2019, Röder et al 2019, Torvanger 2019).
The economics of BECCS are well communicated by the IAM teams, covering the full supply chain from biomass production to the geological storage of CO 2 (table 8). Missing elements however include assumptions on regional availability of financing the for roll out both large scale biomass production (including large scale modern irrigation and fertilisation (Heck et al 2018)), CO 2 transport infrastructure, and assessment and deployment of geological storage.
Probably the most underrepresented and least communicated dimensions that affect the feasibility of BECCS are socio-cultural and institutional/legal. 6. Discussion and conclusions on improving model transparency IAMs have done a tremendous job in offering integrated multi-disciplinary frameworks for discussing plausible climate change mitigation futures. They have been able to both provide and quantify credible narratives of the future. By doing so, they offered a common platform (IPCC 2014) for ongoing discussions on global energy and GHG emission reduction for achieving the Paris Agreement targets. These discussions are vital for policy and investment decisions at global and national scales.
The contribution of this paper is a structured assessment of the transparency of assumptions in IAMs-using the crucial mitigation option of BECCS as a focus. We looked at five particular aspects: biomass availability, bioenergy with carbon capture technologies, CO 2 transport and storage, BECCS costs, and wider modelling assumptions which favour the deployment of BECCS. This is a difficult and timeconsuming task and we employed a traffic light system to communicate levels of transparency (with full methodological details in the appendix). The assessment of transparency also considered parameter inclusion in the modelling framework, i.e. a 'red light' shows that a parameter is not referenced explicitly in the IAM publications we reviewed, and that it is potentially not included in the modelling framework. We took advantage of having one IAM (TIAM-UCL) as our in-house model to allow us to structure the specific model assumptions to investigate. While we disclose all the BECCS relevant data available in TIAM-UCL at the time of writing this paper in the appendix, the colouring of the TIAM-UCL columns of the tables follow the same rules as for the other IAMs and are therefore based only on publicly available journal papers and documentation for TIAM-UCL.
To summarise our findings, we built a transparency ranking system by assigning each colour code a number from 0 to 1, i.e. 0 to red, 0.5 to orange, and 1 to green. Then, for each of the five BECCS aspects investigated here, we calculated individual IAM transparency scores expressed as percentage transparency to full transparency. The results of this exercise are presented in figure 1. Note that this ranking is a snapshot of the status of these models at the time we reviewed them.
A notable finding is that IAMs are transparent in communicating wider system and biomass resource availability assumptions (figure 1). This transparency decreases as we move into modelling details, the least Figure 1. Summary of IAM transparency ranking on BECCS assumptions. The 5 axes represent the investigated BECCS aspects in this work. The percentages on each axis represent transparency of each IAM as percentage of full transparency of modelling assumptions, i.e. 100% means fully transparent assumptions on a given BECCS aspect versus 0% which means no transparency (a parameter is not specified, and potentially it is not included in the modelling framework). transparent assumptions being around the CCS infrastructure. Indeed, all models score over 80% transparency as compared to maximum transparency of the parameters we investigate here when we consider wider system settings such as general discount rate, carbon pricing regime, or availability of other NETs. IMAGE scores the highest, achieving 100% transparency on this aspect. Given the intense recent discussions around biomass availability for bioenergy and BECCS (e.g. Gough 2016, Robledo-Abad et al 2017), the IAMs also score over 60% transparency related to biomass resource assumptions, with REMIND-MAgPIE being the most transparent.
We also found that BECCS cost assumptions are more transparently communicated (between 60% and 80% transparency scores) than technological ones (between 10% and 60% transparency scores). These cost assumptions, combined with a perfect foresight (assuming correct prediction of the future) and a general discount rate usually around 5% (table 9), delay BECCS deployment after the second half of the century. This delay begs two topical questions around how the models account for intergenerational equity and global collaboration for aligning climate mitigation strategies (Lenzi et al 2018), assumptions which we found largely missing from IAMs communication. However, this has now begun to be addressed in recent analyses which vary discount rates (Emmerling et al 2019) or alternatively discuss the explicit intergenerational implications of mitigation pathways with regard to negative emissions technologies (Rogelj et al 2019).
IAMs score over 60% transparency in their assumptions on large scale biomass production. While our 'green labelling' for transparency is assessed from the perspective of biomass availability for use within the global energy system, we recognise that national scale modellers, or readers from other disciplines might wish to see other aspects of land use competition which might not be included in the modelling framework or communicated transparently. The majority of the pathways that are compatible with a SSP2-RCP2.6 future deploy large scale BECCS in the second half of the century. This implicitly assumes that the land will be (i) available, at a time when the demand for land is likely to be high (Obersteiner et al 2018), and (ii) suitable for crop production, which is subject to climate change impacts on land, usually not included in the scenario runs (van Vuuren et al 2017). A further critical assumption is that biomass is supplied without carbon debts. For this assumption to hold, careful temporal carbon accounting with a focus on bioenergy would need to be conducted in each region (Lamers and Juninger 2013). This accounting is not visible in any of the investigated IAMs.
Modelling assumptions around the CCS infrastructure and geological storage were found to have limited transparency with all IAMs scoring below 40%. The IAM community is trying to address this problem either through detailed documentation (e.g. (Luderer et al 2015)), topical model specific studies (e.g. (Muratori et al 2017b) for GCAM), or through model inter-comparisons (notably , which makes technical assumptions visible (parameter values) and explains differences between IAMs). Comparison exercises could be repeated for other technologies, including BECCS for transport fuels and hydrogen. These studies should be complemented by analyses of the influence the technological assumption have on model results. Some IAM inter-comparison studies assess the sensitivity of model results to BECCS technology availability , and CCS assumptions (Koelbl et al 2014). However, for a better understanding of how assumptions influence the model results, cost and technological assumptions should be published with each individual IAM study (Koelbl et al 2014). This is in line with aspirations for the forthcoming IPCC's 6th Assessment report (IPCC 2017).

REMIND documentation
While focusing on assessing BECCS assumptions, we found that it was difficult to separate transparency from completeness (i.e. what the IAMs do not include or is implicit). Our deep transparency analysis in section 4 considered whether the IAM specifically includes the parameter of concern in its modelling framework. We acknowledge that our selection of parameters to investigate is not exhaustive, but is tailored to energy systems modelling needs. Scientists from other disciplines might wish to investigate other parameters which have not been considered here. This could be subject to further transparency analyses. In section 5 we assessed the completeness by contrasting BECCS assumptions in IAMs against the six dimensions of feasibility suggested by the IPCC 1.5°C Special Report (IPCC 2018). We found that IAMs cover fairly well four out of six feasibility dimensions, namely, geophysical, economic, environmental, and technical. What is missing, but critical for establishing BECCS at large scale, are the socio-cultural and institutional-regulatory dimensions. This finding is in line with other studies, e.g. (Gough et al 2018, IPCC 2018).
We do not suggest that IAMs should be expanded to represent these socio-cultural and institutional-regulatory dimensions, but when assessing IAM scenario results it is important to acknowledge the missing elements so other disciplines can participate in the discussion. Some steps in this direction have already been made by IAM researchers recognising the need to complement global results with regional scale analyses to better consider regional specificities of competition for land and its effects on ecosystem services (IPCC 2018).
Overall, we can say that a higher transparency of assumptions in IAMs is possible. Figure 1 shows that for each category we investigated, different IAMs are 'best in class' at communicating transparently in their assumptions. We cannot say that any single IAM is more transparent than the others, but we can say that if desired, higher transparency can be achieved in all the investigated categories. At present, finding modelling assumptions is not straightforward, and requires going from the model documentation to the referenced documents, or to prior model versions for which the documentation is inaccessible. Clear and easy to trace documentation for current and past model iterations would be ideal, so that past results can be understood and differentiated from more recent ones. Some modelling commentators (DeCarolis et al 2017), suggest that model assumptions should be documented with each publication, with links to a data repository. In the particular case of land competition assumptions in IAMs, given the incredible complexity of the topic, huge amounts of data and assumptions for long-term developments which are difficult to assess based on current drivers, increasing transparency through documentation in every publication might be overwhelming for both IAM teams and their readers. In this case, increased transparency could be achieved through multi-disciplinary workshops in which specific assumptions are discussed in specific contexts from a multitude of angles .
One alternative to help increase transparency is the provision of open-source models, which GCAM and MESSAGEix-GLOBIOM teams already do. While they do provide training with their models, they remain complex, and running them with full understanding of underlying assumptions and drivers is a very time-consuming task. A more meaningful approach to increase transparency could consist in iterations with different audiences, gradually opening to scrutiny other assumptions in specific contexts (Strachan et al 2016).
A final key element is building explicit resources into projects for transparency work. In practice this is difficult to achieve, as the funding for model maintenance tasks is intermittent or inexistent, and the time and reward of researchers comes from high profile publications (Strachan et al 2016). But this brings us full circle to increase the transparency of assumptions in IAMs as a bridge to funders, policy makers and other disciplines (DeCarolis et al 2017). This would be a timely and critical exercise to increase the recognition of IAM results, and to enable different communities to work together for climate mitigation.