Improving life cycle assessment for carbon capture and circular product systems

Along the years many guidelines and frameworks were reported by specialists. The following sections shows a non-exhaustive presentation of the most recent/established guidelines. Table 1 summarises the key modelling aspects of some common and recent guidelines that have been used for industrial and academic assessments.

LCA can be used to identify critical points during the design of novel manufacturing processes by comparing them with commercial products/processes (Langhorst et al. 2022). Multiple scenarios can be considered through the system operating from small to large operational scale (Cucurachi et al. 2018). At the early stage of the technology development, designers can create long-term strategies to reduce emissions and compare different scenarios to evaluate the environmental performance of different products, operational conditions and processing strategies. However, the comparison of LCIA of CCUS processes can produce misleading results due to the lack of representative data to build the mass and energy inventories, resulting in high uncertainty and misleading decisions (Moni et al. 2020).

The reproducibility of an LCA or even the re-use of technology information is limited by the lack of mass and energy datasets or enough details of the process’ mechanisms, kinetics or estimates. Alongside, lack of transparency in the LCI, whether deliberately (e.g. to protect commercially sensitive information) or not, makes replication of the study challenging and in turn limits the future applications of the LCA in question such as using it as an upstream process in another LCA. This leads to an overall reduction in the quality of the study (Bisinella et al. 2021). Quality examples of this are Blume et al. (2022) where the authors provide the material flows required to build an industrial-scale battery in a concise flow chart and Sharma et al. (2021) who provide the inventory in tables to identify material and energy flows for different foreground processes of a pharmaceutical manufacturing process. The comparison of the environmental performance is especially important for emerging technologies to produce commodities since they should provide high beneficial impacts to be considered as an alternative to the current production process. Although the comparison of the assessments and/or comparative assertions are the focus during the scope definition phase for all CCUS frameworks (EC 2017; ICCA 2022b; Langhorst et al. 2022), a variety of LCA modelling options are available, also hampering the comparability of LCIA of different studies.

The considerations of specific LCA guidelines vary at different levels, which includes fundamentals adjustments. Assumptions and estimates during the LCA modelling start at goal and scope definition, in which not only reference flows and functional unit are defined but also system boundaries, cut-offs, spatial–temporal considerations, co-products and waste management strategies might be established. The complexity of the model and the number of assumptions required can increase substantially as the model becomes circular, accounting for the reuse and recycling integrated with carbon capture and waste management under temporal, regional and supply–demand constraints. Indeed, the application and/or comparison of different LCA is not a straightforward task and could be challenging due to conflicting outcomes that depend on primary assumptions and interpretations. ‘Traditional’ LCAs are modelled with a linear approach, in which some cut offs are applied in the system boundary to determine the environmental outcomes of a specific manufacturing process or product. Qualitatively, the linearity of products manufacturing is used to simplify the complex ‘carbon web’ of the circular economy, which was represented by the diagram from Fig. 2.

Other key-decisions could diverge significantly, such as the allocation of burdens and benefits due to the implementation of circular strategies.

Frameworks for the best practice of LCA for both industrial processes and emerged technologies were reviewed and the key-aspects of its application on the context of a circular economy were highlighted. Temporal and regional aspects were identified as key modelling considerations that are not clearly covered on former guidelines, then their practical modelling was critically reviewed in PART II. Moreover, LCIA uncertainties and the system sensitivity under process can be decisive drivers for technologies screening on an industrial decarbonisation scope. Both uncertainty and sensitivity are strongly associated to the quality of data that is used for LCA modelling, as discussed in PART III.

Temporal aspects reflect the complexity of the circular economy; if correctly accounted, they could provide important information related to delayed emissions, and recycled products over time, helping to build management and mitigation strategies to avoid depletion of natural resources. The gradual accountability of captured emissions over the product lifetime are required to explain CCUS systems and its modelling can be complex (Beloin-Saint-Pierre et al. 2020; Lueddeckens et al. 2020). Then, simplifications are settled by the guidelines (such as those in Table 1) to facilitate both LCA modelling and LCIA interpretation. For comparison purposes, a straightforward and simple LCIA evaluation is possible by aggregating emissions over time or neglecting some system dynamics. However, the loss of time resolution could lead to misleading conclusions, which could hamper the development of a circular economy based on CCUS. In this context, policy makers and stake holders might identify the emissions profile along the years (with high temporal resolution) to meet policies targets and/or carefully address the future generations impacts.

Circular systems consider the valorisation of alternative carbon sources to avoid emissions in the production of materials and products that would be otherwise produced from non-renewable resources, preventing with this the release of greenhouse gases (CO_{2 eq.}) into the atmosphere (ICCA 2022b). The accountability of these factors is dynamic and then, strongly related to temporal assumptions on LCA. According to the most frequent guidelines (EC 2017; Langhorst et al. 2022; Zampori and Pant 2019), the amount and duration of carbon storage should be reported as additional information at LCIA. Following the LCA principals from ISO 14067 (2018), to calculate the climate change impact, the discounting of emissions over time and temporary/permanent carbon storage and/or delayed emissions shall not be included. Another approach is to consider the avoided emissions as additional to or independent of climate change indicators (Langhorst et al. 2022) and to assume that it does not contribute to climate change during a specific period. The lack of temporal resolution refers to the assumption that all input and output emissions are released at the present time (EC 2017; ICCA 2022a; Langhorst et al. 2022) by averaging the characterisation factors scores over 20, 100 or 500 years—as suggested by the IPCC guidelines. On bioenergy systems, for instance, the delayed emission of the CO₂ into the atmosphere is often relatively short due to their life cycle which is often between 3 and 30 years depending on crop type. Then, the advantage of the renewable sources for energy production relies on the carbon-savings compared to the fossil fuels options. Under the Langhorst et al. (2022) guidelines, the comparison of fuels and energy storage systems shall include each temporary storage on their cradle-to-grave LCIA; no significant effect is expected. It is important to highlight that the modelling of biofuels production provides a straightforward analysis of the whole life cycle of the final product since its final use/disposal at the EoL is not as complex as for other chemicals as in general, the carbon is released at point of use. In the latter case, different applications/uses are available to transform chemicals into new products that regards on different EoL modelling for each potential application.

The temporary or permanent carbon storage could vary according to different waste management strategies. Waste-products could be valorised by recycling or refurbishing at their EoL. Materials such as steel, aluminium or polymers can be produced from virgin, recycled or recovered mixed materials, causing or not changes on the chemical property of the final product. For a cradle-to-grave approach, the temporary storage is required for comparing final products with different composition (chemicals, material and others). Emission time profiles may be recommended as additional data by other guidelines (ICCA 2022b; Langhorst et al. 2022), but no standard practice is provided. Langhorst et al. (2022) discussed the delayed benefits of CCU and CCS. The CCU affects the emission time profile just if the conventional counterparts are replaced by CO₂-based products. For CCS technologies, the EoL emission never occurs, which mathematically means permanent storage with zero emission.

In fact, intrinsic differences of the systems might include temporal decision on the LCA in diverse levels by establishing short- or long-term analysis, dynamic function units, different time horizons for the emissions, as well as covering the period of data validity or technological obsolescence. The indirect factors largely relate to the source, deployment and current coverage of the technology, as well as the age and source of data. Figure 5 shows a quick-reference for the key-temporal aspects and tools for LCA application. Considering the complexity of dynamic modelling and the lack of standards to its practice, an overall of the temporal modelling in LCA and the available tools were briefly reviewed in the following sections.

The transition towards a circular economy can only take place over a specific time frame, thus high temporal resolution on LCA modelling is essential. The lack of a precise temporal definition partly derives from the lack of consensus on how to define temporal estimates and dynamic methods (Beloin-Saint-Pierre et al. 2020). The definition of time on a LCA starts on the goal and scope phase. The first decision regards to Time Horizon (TH) of a consistent system boundary. The TH could be associated to the length of the life cycle of the product or service, to the inventory modelling, and, finally, to the TH of the characterisation factor, as shown in Fig. 6.

The temporal estimates and considerations are often related to LCA drawbacks because they are usually lost during the inventory calculation (if the static approach is used) or during the impact assessment where the potential damages/gains can be aggregated in a specific time (Levasseur et al. 2010). In a static approach, for instance, we can assume that all emissions are taking place at the time where the evaluation of the potential environmental drawbacks and benefits is made by aggregating emissions at one point (Langhorst et al. 2022; Levasseur et al. 2010). Time can be included on LCA directly by dynamic calculations where input and outputs flows change over time and gradual impact of emissions are estimated by dynamic characterisation factors (CF). Figure 7 summarises the conceptual representation of the LCIA outcomes for both static and dynamic approaches.

Adapted from Pigné et al. (2019) and Langhorst et al. (2022).Indirect temporal aspects could be related to technology development and deployment, age of data sources and lifetime of the project. The time could also appear indirectly and limit the systems analysis by their boundaries. In this case, the end-of- life of one product can be settled outside the boundaries by using a cradle-to-gate approach, which consider just part of the emissions of the life cycle of a product. A cradle-to-gate approach is usually used to compare difference environmental performance due to changes in the production processes (e.g. efficiency or technologies) of the same products (Fernández-González et al. 2022; Pinto et al. 2022). Other cut-off approach is related to the waste processing, where just the first grave (recycled content) is accounted on the LCA modelling (ISO 14044 2006). Clearly, the real benefits over time of additional recycling are lost, leading to underestimation of environmental gains.

The inclusion of technosphere and biosphere exchanges with a CLCA, dynamic CFs or inventory increases the time representativeness of the LCA. These estimates/models could be combined or not, which means that the study can consider the temporal aspects of the inventory with or without dynamic CFs. To reduce the model complexity of the inventory resolution, Beloin-Saint-Pierre et al. (Beloin-Saint-Pierre et al. 2014) suggests that it is not necessary that the whole inventory to be dynamic, especially if there are very small emissions. On the scope of the CFs, the TH is important due to the temporal delay of several years between releases and the effective start of the potential environmental damage. The greenhouse gas emissions, for instance, show different behaviour depending on the TH considered and the IPCC provides estimates of 25, 100 and 500 years for the global warming potential (GWP). On the perspective of the LCA practice, 25 years would obviously be too short, underestimating the potential impacts of CO₂ emissions when compared to CH₄ on the GWP, which are 72 times higher than the former in a 20-year TH (Guest et al. 2012; Levasseur et al. 2012). The afore mentioned difference is reduced by 25 times in a 100-year TH, which is the most usual timeframe for GWP (Langhorst et al. 2022). Moreover, the GWP-500 can be too long leading to high complexity and uncertainty scenarios, as well, ignore the urgency of environmental problems, as an excuse for no action. Clearly, the scope has a key role for settling this timeframe in all situations and the GWP-500 are still demanded for inferring how the current technologies/strategies will impact the future generations survival, for instance. On a dynamic approach, the correct TH should be chosen carefully to avoid misleading results due to resolution loses (Levasseur et al. 2010).

The temporal aspects could be used to improve the time resolution by accounting the cycle in which the CO₂ is captured during the growth of crops and released after years to produce bioenergy or on the EoL of different products (Guest et al. 2012). For energy crops, the carbon storage in the technosphere benefits are far less pronounced for low rotation periods and high TH (Guest et al. 2012). Almeida et al. (2015) concluded that the increase of the DLCA model complexity did not add value for the analysis of GWP emissions during the energy production from perennial energy crops–short-lived product. The authors also suggested that the period of temporary carbon sequestration in the biomass is short and immediate metrics (e.g. IPCC GWP) should be used (Almeida et al. 2015). It is worth to mention that, traditionally, the impact over time is neglected and the use of biomass as a renewable source is discounted (biogenic contribution) by the difference between the carbon that is captured and the carbon that is released during the harvesting or at the EoL of the product (SETAC 2011).

Briefly, DLCA allows for an increase in temporal resolution in several aspects that include (but are not limited to) the definition of the functional unit (e.g. one year of energy consumption), the temporal distribution of elementary flows over time (foreground and/or background processes—supply chain (Aldaco et al. 2019) and marked demand (Ryu et al. 2022)), datasets (Sacchi et al. 2022), technology deployment (Faber et al. 2022) and dynamic CFs (Beloin-Saint-Pierre et al. 2020; Lueddeckens et al. 2020). To achieve the goals of industrial decarbonisation, a specific TH might be set by government policy. In this sense, increasing the temporal resolution at LCA will enable a clear understanding of the carbon capture efficiency of different CCU and CCS technologies.

A Generalised structure for regional LCA has been provided by Yang and Heijungs (2016). Their method allows an analyst to determine the regionalised impacts (\({h}_{r}\)) based on the production of input materials from other regions, which is defined by the next equation:

For hr, each element of the equation is defined by the next expressions

Environmental matrix:

Technology matrix:

Functional unit vector:

Note that each of the elements of \({h}_{r}\), \(\mathbf{C}\) _r, \({\mathbf{B}}_{r}\), \({\mathbf{A}}_{r}\) and \({f}_{r}\) is a vector or matrix representing \(j\) regions. Even though \({\mathbf{B}}_{r}\) might appear square in the generalised form shown above the constituent matrices, \({\mathbf{B}}^{11}\) to \({\mathbf{B}}^{jj}\) may be non-square, resulting in \({\mathbf{B}}_{r}\) being non-square. By applying this methodology to LCA, analysts are able to assess the impacts of a product or process in different regions with greater precision based on local impacts such as manufacturing trends and electricity grid mixes. LCA databases such as Ecoinvent contain regionalised datasets for the majority of their processes making this method widely implemented. However, the regions used in the datasets are often broad groupings of countries and geographies such as Europe (designated as ‘RER’ in Ecoinvent) which is not technologically, geographically or ecologically homogeneous.

Regional scales vary between LCI and LCIA which is one of the major challenges in regionalised LCAs due to each being based at different regional scales (Mutel et al. 2012). For example, LCI data may be based on political borders between countries, states and regions, whereas LCIA scales are often derived from ecological factors such as population density, water courses.

The most commonly applied regionalisation method is the regional outputs approach (ROP) described by Yang (2016) which uses the basic matrix representation of LCIA with some modifications:where \(\mathbf{A}\) is the technology matrix and a column represents a process and a row a product, \(\mathbf{B}\) is the environmental matrix that represents the quantity of emissions or natural resources emitted/consumed by the processes in \(\mathbf{A}\), \(f\) is the final demand column vector related to the functional unit, \(R\) is the regional output percentages, \(T\) transposes \(R\) and \({h}_{r}\) consists of row vectors where each row is the quantity of a products life cycle emissions that occur in different regions. If \(R\) was representative of three regions contributing to the production of 10 kg of product with 10 kg from region 1, 8 kg from region 2 and 2 kg from region 3 it would be of the form:

More complex and spatially accurate models exist that use Input–Output based approaches such as the interregional input output model (IRIO) also described by Yang (2016):where \({\mathbf{B}}^{i}\) denotes environmental emissions from process in region \(i\), \({{\text{A}}}^{ij}\) denotes commodity flows from processes in region \(i\) to processes in region \(j\). \({f}^{i}\) denotes final demand in region \(i\).

A hybrid approach combining IRIO and ROP models is also available:where \({R}^{i}\)(\(i\) is 1, 2, …, \(n\)) contains regional output percentages for the subregions defined within region \(i\). The ROP and hybrid approaches provide greater granularity than the method in Eq. 1, albeit with a larger input data requirement and computing time. Therefore, the application of these methods may not be suitable when running multiple LCAs such as when conducting Monte Carlo analysis.

Regionalisation has been implemented into most of the commercially available LCA software. It is generally handled in two ways: flow-based or geo-spatially based. Flow-based approaches use elementary flows which are extended to include regional information whereas the geo-spatial approach uses GIS data to derive LCI datasets.

One alternative for increasing the temporal resolution on the LCA is addressing dynamic conditions in both the inventory and the LCIA phases with the Temporalis software tool. It supports 3 types of CFs that could be static (which means no changes over time), dynamic (where the values change over time and the impact still occur along emissions) and extended (which provides the decays rates of each CF over time) (Cardellini et al. 2018). In all approaches of this methodology, no carbon neutrality is assumed for biogenic carbon, which means its emission is accounted immediately. The dynamic impact assessment method spreads the environmental impact over time and a best-first exploratory strategy is applied to organise them as function of their priority (Cardellini et al. 2018; Zhang and Korf 1993). Then, the technosphere exchanges that contribute the most receive the highest priority and those with low relevance are neglected, as a cut-off strategy. The time of occurrence of all environmental interventions is added to a timeline satisfying the requirement of the DLCA even when the static CF are selected. The Temporalis methodology is efficient in increasing the dynamic resolution of both inventory and dynamic characterisation of emissions. The main advantage of the Temporalis approach remains on an open source generic method and tool for DLCA implementation available in Brigthway2 with framework (Cardellini et al. 2018).

Other exploratory methods are available on the literature to account the temporal aspects on LCA (Beloin-Saint-Pierre et al. 2014; Tiruta-Barna et al. 2016), but with some software limitations on running time and memory requirements. Methodological improvements considering the impact propagation over time and additional background data were suggested by (Pigné et al. 2019). The dynamic outcome of this approach considers a clear reference to process and supply chain functioning, as well as a proper temporal database linked to the background LCI (Pigné et al. 2019). Another tool software is available online for testing purposes (DyPLCA), but the access to the temporal database is protected by property rights (Pigné et al. 2019).

Temporally differentiated life cycle inventories and time-dependent, or at least TH-dependent, characterisation improve the accuracy of the LCA. DLCA are attractive for companies because the calculation results are more accurate and often lower than the static ones. However, the practical use of the former methods and tools presents barriers for their implementation in different contexts of the framework application since they are not yet available as user-friendly platforms.

Related to different perspectives, although the temporal representativeness of the supply chain have been modelled by the DyPLCA tool (Pigné et al. 2019), the fluctuation on the market demand have not been considered during the modelling. Moreover, the existing market and surrounding product systems are not considered on the LCI of static models because they operate under steady-state conditions where the average demand and supply is generally covered in any scale (Marvuglia et al. 2013). From an industrial decarbonisation perspective, different scenarios with this scope could support the decision making process for complex systems (Marvuglia et al. 2013). The literature suggests the application of prospective modelling, where the CLCA is usually used on their calculations to project potential environmental impacts to the future and cover the interconnectivity between product system -environment. The fundamental differences on modelling can be used to classify the prospective approach in three different categories: economic models, historic trends and future scenarios (Beloin-Saint-Pierre et al. 2020). Albers et al. (2019) applied this approach to evaluate consequences of policy-driven transport strategies with time-sensitive supply–demand and dynamic CF (C_biogenic). Aldaco et al. (2019) used the prospective approach to compare the environmental outcomes of formic acid production from carbon capture and storage CCS (geological) by evaluating the potential evolution of the energy’s supply system under different climate change scenarios (Aldaco et al. 2019); nevertheless, the interference of the final product on the market was not accounted by this approach. At the best of our knowledge, any tool for the application of the prospective modelling mentioned here before is available. Furthermore, the literature presents applications that are not limited to CCS and CCU (Beloin-Saint-Pierre et al. 2020), which are beyond the scope of this paper.

The inherent variations of the systems could consider time within their scope by setting short- or long-term analysis, dynamic function units and TH, and also by covering the period of validity of the data or technology obsolescence; nevertheless, the inclusion of temporal aspects in LCIA could have direct or indirect affectations to it. The indirect aspects are related mainly to the age and source of the data that is used, the technology deployment and the up-to-date coverage. Related to direct affectations, decisions regarding to LCA methods (e.g. consequential), dynamic or static inventories and CFs affect the analysis of the environmental outputs and can be assessed by different programming tools.

The complexity of the modelling, the availability of dynamic data, and the tools utilised during the study hampers the temporal calculation on LCA. Only the DyPLCA and Temporalis frameworks are available as tools. Furthermore, a limited selection of environmental categories has dynamic CFs. The process network modelling is complex due to the interaction of extensive timespans of the foreground (analysed process) and background (all supply-chain) data (Pigné et al. 2019). The temporal uncertainty and model complexity must be reduced by choosing the best compromise between accuracy and computational practicality (Pigné et al. 2019). Although the DLCA outcome provides high-quality information, its interpretation could not be as straight forward as the static approach. In this sense, the LCIA’s scope, the technological state of the art, and the accessibility of the data should all be carefully considered when evaluating the inclusion of temporal aspects.

Regionalisation requires regional inventories and CFs which can be difficult to obtain at the same regional resolution. Given this, the onus of improving regionalisation relies on data collectors to improve the resolution of regional data and regional CFs, although this may be a monumental task involving the geopolitics of regions as regional CFs are often provided by state bodies. This can lead to different impacts when considering the regional impact of a process operating near and trade over international and intranational borders.

When circularity and regionalisation are considered in tandem, the processes remain the same. However, the regional vector or matrix (depending on the approach) needs to be adjusted to include a greater number of product transports between sub-regions leading to a more demanding analysis that also require temporal analysis as ‘wastes’ are generated and consumed cyclically.

In this section, we aim to address the shortcomings of the ISO guidelines with regards to data quality analysis and provide LCA analysts with methodologies to conduct UA and SA to improve the quality of their LCAs. All input data to an LCA model has some degree of uncertainty. Whether that uncertainty arises during experimental data collection, extrapolating data through time and across geographical boundaries, unverified data or from many other sources, it must be accounted for in the LCA output to accurately contextualise the importance of the results. Sensitivity analysis (SA) is then applied to the uncertainty analysis (UA) to model how the system responds to the uncertain values in the system. Properly applied, uncertainty and sensitivity analyses provide information on the predicted distribution of outputs and how a system reacts to perturbations in the inputs from a base case (Cucurachi et al. 2021).

Considering the environmental impacts of emerging technologies, it is essential to evaluate both the uncertainty associated with the technology and the sensitivity of the system to perturbations in the inputs. However, uncertainty is rarely included in published LCA studies. A meta-analysis of papers published in leading LCA journals conducted by Bamber et al. (2019) found that only 19% of ALCA and 15% of CLCA studies included some form of UA. Lo Piano and Benini (2022) found similar results as well, identifying that many authors confuse UA with SA or conduct only UA or SA, but rarely both. Most major LCA software have built in UA, SA, Monte Carlo simulations (MC) for UA, and one-at-a-time (OAT) SA (Igos et al. 2018). Bamber et al. (2019) attribute the absence of UA to several factors: a lack of understanding of the importance of UA; the time required to conduct a thorough UA; and insufficient uncertainty data available to the analyst. Additionally, the amount of information presented in the ISO 14040 and 14,044 guidelines on data quality analysis is scarce, leaving the analyst to conduct UA and SA without any specific guidance as to which methods are appropriate; therefore, many analysts choose not to conduct UA or SA.

SA is the process of determining how the uncertainty in the inputs affects the output uncertainty of a process, model or system. SA is generally conducted by recalculating outputs with small changes to two or more inputs and quantifying the impact that each varied input causes, in contrast to UA, which does not identify the assumptions responsible for uncertainty in the output (Saltelli et al. 2019). Similarly, ISO 14044 (2006) suggests to LCA analysts to conduct SA as part of their analysis but do not give a definitive method to do so.

The development of new LCA studies including both an UA and a SA is crucial to determine the environmental impacts of new and emerging technologies. Doing so will provide LCA analysts and policy makers with more information to understand the system as it develops or changes through time, as well as for the systems designers to visualise the contributors to environmental impact and the mitigation strategies. The framework developed here for UA and SA is presented in Fig. 9. Following these four steps ensures that the data quality analysis is complete and consistent with the state of the art described in reviews of uncertainty and sensitivity in LCA such as Lo Piano and Benini (2022), Bamber et al. (2019) and Igos et al. (2018). We also encourage LCA analysts to perform uncertainty considering more sources of uncertainty to better represent their systems, and to allow readers and decision makers to better understand the applicability of LCA studies.

Global SA methods should be considered by LCA analysts as the first choice for SA due to the quality of the analysis being higher than that of local methods. Local SA should be reserved for very simple processes with few inputs. Therefore, implementation of more robust SA methods should be a consideration of LCA software developers, such as Sobol’ indices and squared Spearman correlation coefficient which were determined to be the most appropriate for LCA by Groen et al. (2016). Journal editors, and reviewers should consider the appropriateness of the sensitivity methods applied in LCA studies before approving manuscripts for publication.