Angelica Mendoza Beltran,Brian Cox,Chris Mutel,Detlef P. van Vuuren,David Font Vivanco,Sebastiaan Deetman,Oreane Y. Edelenbosch,Jeroen Guinée,Arnold Tukker
First published: 21 November 2018 https://doi.org/10.1111/jiec.12825

Summary

Prospective life cycle assessment (LCA) needs to deal with the large epistemological uncertainty about the future to support more robust future environmental impact assessments of technologies. This study proposes a novel approach that systematically changes the background processes in a prospective LCA based on scenarios of an integrated assessment model (IAM), the IMAGE model. Consistent worldwide scenarios from IMAGE are evaluated in the life cycle inventory using ecoinvent v3.3. To test the approach, only the electricity sector was changed in a prospective LCA of an internal combustion engine vehicle (ICEV) and an electric vehicle (EV) using six baseline and mitigation climate scenarios until 2050. This case study shows that changes in the electricity background can be very important for the environmental impacts of EV. Also, the approach demonstrates that the relative environmental performance of EV and ICEV over time is more complex and multifaceted than previously assumed. Uncertainty due to future developments manifests in different impacts depending on the product (EV or ICEV), the impact category, and the scenario and year considered. More robust prospective LCAs can be achieved, particularly for emerging technologies, by expanding this approach to other economic sectors beyond electricity background changes and mobility applications as well as by including uncertainty and changes in foreground parameters. A more systematic and structured composition of future inventory databases driven by IAM scenarios helps to acknowledge epistemological uncertainty and to increase the temporal consistency of foreground and background systems in LCAs of emerging technologies.

Introduction

A robust assessment of the environmental impacts of product systems is the basis for assertive policy, business, and consumer decision making (Hellweg and Canals 2014). Life cycle assessment (LCA) has developed into an environmental decision-support tool to assess product systems. Some LCAs, however, refer to product systems that either do not yet exist, that are not commercially available, or that refer to decisions about the future. These forward-looking applications of LCA, or so-called prospective LCA (in line with the definitions of Arvidsson and colleagues [2017] and Pesonen and colleagues [2000]), are thought to help in anticipating unintended consequences of future product systems and to support environmentally assertive product design (Miller and Keoleian 2015). Prospective LCA has proven to be valuable in a range of cases, from assessing future public policies (Dandres et al. 2014, 2012) and emerging technologies (Arvidsson et al. 2017; Frischknecht et al. 2009) to the analysis of future production and consumption systems (Van der Voet et al. 2018). Nonetheless, in addition to dealing with the uncertainty related to any complex system (ontic uncertainty), prospective LCAs suffer from a particular type of epistemological uncertainty, that is, uncertainty “that arises when future systems are modelled, because the future is inherently uncertain” (Björklund 2002, 65). Addressing epistemological uncertainty is therefore a crucial challenge in the development of prospective LCAs.

A common approach for dealing with epistemological uncertainty in prospective LCAs is to integrate future scenarios (Pesonen et al. 2000; Spielmann et al. 2005). In this study we use the following definition of scenario: “… a description of a possible future situation relevant for specific LCA applications, based on specific assumptions about the future, and (when relevant) also including the presentation of the development from the present to the future” (Pesonen et al. 2000, 21). Common approaches to integrating scenarios in prospective LCA draw from multiple databases exogenous to LCA to address future sociotechnical changes or so-called exogenous system changes (Miller and Keoleian 2015). For example, the New Energy Externalities Developments for Sustainability (NEEDS) project (NEEDS 2009) modeled the future supply of metals, nonmetallic minerals, electricity, and transport using different scenarios at various levels of optimism regarding technological improvements, cost reductions, and market growth rates. NEEDS and other external databases, such as the IEA (International Energy Agency 2010), were used in the Technology Hybridized Environmental-Economic Model with Integrated Scenarios (THEMIS) (Gibon et al. 2015) to integrate future changes in electricity production, industrial processes, and climate change mitigation policies into a hybrid input-output (IO) LCA model (Bergesen et al. 2014, 2016; Beucker et al. 2016; Hertwich et al. 2015). Another example is macro-LCA (Dandres et al. 2012), which combined LCA with future changes in economic structure and energy production based on computable general and partial equilibrium models, respectively. Finally, Van der Voet et al. (2018) identified important supply-related variables that are likely to change in the future of metal production (e.g., technologies’ shares of production, resource grade, and efficiencies of technologies), and then adapted these using various assumptions and external data sources.

While the above examples are valuable for prospective LCA, they suffer from limitations. A first limitation is that the development of future scenarios is often inconsistent and lacks transparency. Scenario development involves two steps: scenario generation and scenario evaluation (Fukushima and Hirao 2002). Scenario generation refers to the formulation of assumptions about the future, while scenario evaluation refers to the assessment of such assumptions during the LCA phases, especially the life cycle inventory (LCI) phase and the life cycle impact assessment (LCIA) phase (Fukushima and Hirao 2002). Because scenario generation and scenario evaluation are often mixed, it is difficult to establish which inventory parameters have been changed and, most importantly, to discern whether future assumptions are coherent among technologies, economic sectors, and regions (consistent changes). Part of this issue arises from the use of different datasets as sources of scenario information, a procedure that increases inherent uncertainties (Gibon et al. 2015) and makes the process of scenario generation possibly unharmonized. Another limitation is that technology maturity (e.g., penetration and efficiency) is often not accounted for, thus misrepresenting future technology mixes (Dandres et al. 2012). Moreover, because technological development is intertwined with both economic development and predictions of product technology-supply mixes, such relationships should be appropriately reflected in a scenario covering all economic sectors worldwide. Finally, the reproducibility of some approaches can be hampered by the large amount of required data and the difficulty to trace the assumptions that were made during the scenario generation.

To overcome the above limitations for scenario development in prospective LCA, we first propose to explicitly differentiate between scenario generation and scenario evaluation. For scenario generation, we propose the use of system-wide integrated assessment models (IAMs) as a platform for calculations of consistent, worldwide scenarios covering all economic sectors. IAM scenarios are possible socioeconomic and technological pathways of future development (van Vuuren et al. 2014) that can help explore different futures in the context of fundamental future uncertainties (Riahi et al. 2017). Masanet and colleagues (2013), Plevin (2016), and Pauliuk and colleagues (2017) highlight the unrealized potential of IAM scenarios as consistent sources of information for prospective assessments.

For scenario evaluation, we introduce a novel approach that systematically integrates the scenario information of the technology-rich IAM Integrated Model to Assess the Global Environment (IMAGE) (Stehfest et al. 2014) with one of the most broadly used life cycle inventory databases in the LCA community, the ecoinvent database (Wernet et al. 2016). In contrast to the recent work of Arvesen et al. (2018) and Pehl et al. (2017), we concentrate on evaluating the usefulness of IAMs for prospective LCA rather than on informing the IAM with the prospective LCA results. Our approach can thus be understood as an alternative opportunity to further reconcile the knowledge from the IAM and the LCA communities (Creutzig et al. 2012) that now hold different views on how to perform future environmental impact assessments.

The main research question of this study was as follows: “How can IAM scenarios be systematically linked with LCI parameters to account for future changes in prospective LCAs?” To answer this question, we focused on a case study comparing the relative environmental impacts of two mobility alternatives in the future. Despite this focus, the utility of the proposed prospective LCA approach, for instance, for emerging technology LCA (ETLCA), is expected to be beyond the transportation sector. We believe that this wider, structural utility can be realized by linking all sectors available in IAM scenarios with LCI parameters. However, we did not choose such an ambitious scope because each sector has its own peculiarities and complexities, and we first needed a proof-of-concept for just one sector.

Electric vehicles (EVs) and internal combustion engine vehicles (ICEVs) are compared, given that future changes play a key role in the impacts of these two mobility alternatives. Drawing from previous research, we focused on changes in the electricity sector. Specifically, the relative carbon footprint of EVs is highly influenced by the electricity mix (Cox et al. 2018; Bauer et al. 2015), and extreme cases can lead to counterintuitive results; for instance, in Australia, the prevalence of coal power causes EV to underperform (Wolfram and Wiedmann 2017). Our approach can thus address a range of questions posed by different stakeholders, such as vehicle producers, who might be interested in the question, “What will be the environmental impacts of EVs in 2050 and what are their key drivers?”; and policy makers, who might be interested in the question, “Will a transition to EVs in the future bring environmental benefits?” Finally, we contribute to the integration of knowledge from the IAM and LCA communities, with the aim to increase the robustness of prospective LCA assessments, by linking macro scenarios into the micro- or product-level LCA (Guinée et al. 2011).

Methods

We first present an overview of the proposed approach. Next, we provide detailed insights into how scenarios are generated using IAMs and particularly IMAGE. Next, we present the Wurst software, which is the tool developed to adapt the LCI background data using the IMAGE scenarios as a source of information. Finally, we describe the case study and the scenarios used in the case study.

Approach Overview

This study presents a novel approach to introducing consistent and systematic future changes in a prospective LCA application to calculate more robust prospective results (see figure 1 for an overview). Such changes refer to the LCA background system, namely, those processes and emissions that are part of the supply chain of the studied product system, for example, the electricity mix used to charge and produce EV batteries. This means that indirect emissions are accounted for. In addition and in line with a full life cycle approach, direct emissions are accounted for but are left unchanged in the foreground system. In particular, despite the long-term focus of the study, no changes have been made to the processes, emissions, and parameters describing the product itself, for example, vehicle energy use, vehicle size, lifetime, driving patterns, and battery size. These parameters have been found to contribute to the variability of future EVs, but the largest contributor to variability is electricity used for charging (Cox et al. 2018). We keep the EV and ICEV foreground unchanged to focus on the background changes. Following Fukushima and Hirao (2002), we developed scenarios in two steps: (1) scenario generation and (2) scenario evaluation.

Scenario generation: This step refers to the process of scenario formulation and calculation. The IAM model IMAGE (Stehfest et al. 2014) was selected as the modeling framework used to generate consistent scenarios. IMAGE was selected due to its wide coverage of world regions, technologies, and economic sectors as well as its range of scenarios that are key to addressing uncertainty. The following paragraphs provide descriptions of the IMAGE model, the type of scenarios developed by the model, and the specific scenarios used in the case study.
Scenario evaluation: This step refers to the assessment of the scenarios in all the phases of LCA. Yet, in this study, particular attention is paid to the evaluation of scenarios in the life cycle inventory phase. We identified three steps needed to accomplish this: first, analyzing the background system to identify the inventory parameters (i.e., input and output flows as well as processes) that are affected by future changes; second, adapting these parameters using information from the IAM scenarios; third, using the adapted inventories to calculate the prospective LCA results of specific products.

Figure 1. Overview of the proposed method for scenario development in prospective life cycle assessment (adapted from Fukushima and Hirao [2002]) using the IMAGE 3.0 framework (http://models.pbl.nl/image/index.php/Framework_overview) as an integrated assessment model (IAM)

Relevant inventory parameters were adapted using so-called cornerstone scenarios (Spielmann et al. 2005), as these scenarios refer to either unknown or new future situations for all parameters together. These scenarios have been chosen, as they better inform long-term and strategic decision making, which are fundamental characteristics of prospective LCA. The alternative is to use “what-if” scenarios, which test changes in specific parameters to compare well-known alternatives in a sensitivity fashion (Pesonen et al. 2000). However, we did not choose this option, as it is less structural than cornerstone scenarios because changes of only few parameters are captured. The approach of this study is distinct from other implementations of cornerstone scenarios (Spielmann et al. 2005) as we derived future changes of relevant parameters from the IAM-based scenarios instead of making separate assumptions for each parameter and then combining them. We developed and applied the Wurst model (v. 0.1) in this study (https://wurst.readthedocs.io/index.html) for the parameter identification and adaption steps, as will be described in detail below. The LCA results of EVs and ICEVs were calculated with the Brightway2 (v. 2.1.1) software (Mutel 2017).

Scenario Generation: Using IMAGE to Develop Scenarios

We used the IAM IMAGE 3.0 to generate scenarios (for a detailed model description, see Stehfest et al. 2014). In general, IAMs have been developed to describe the relationships between humans (the human systems) and the natural environment (the Earth system) and the impacts of these relationships that lead to global environmental problems, such as climate change and land use change. IAMs build on functional relationships between activities such as the provision of food, water, and energy and their associated environmental impacts. The human system in IMAGE includes economic and physical models of the global agricultural and energy systems. The Earth system includes a relatively detailed description of the biophysical terrestrial, ocean, and atmosphere processes.

Because this study focuses on the electricity sector, we will briefly describe the energy model of IMAGE, “The Image Energy Regional Model” (TIMER) (de Vries et al. 2001; van Vuuren 2007). TIMER consists of a technical description of the physical flows of energy from primary resources through conversion processes, transport systems, and distribution networks to meeting specific demands for energy carriers or energy services. The model determines market shares for energy technologies based on the costs of competing technologies. It includes fossil fuels and renewable or alternative sources of energy to meet the demand, which depends on population size, efficiency developments, income levels, and assumptions on lifestyle. The model generates scenarios for future energy intensity and fuel costs, including competing nonfossil supply technologies. It models emission mitigation through the price signal of a carbon tax that induces additional investments in more efficient and nonfossil technologies, bioenergy, nuclear, and carbon capture and storage, thus changing market shares of different technologies. In this way, the TIMER model allows the generation of both baseline and mitigation energy scenarios as part of broader IMAGE scenarios, both of which are used to inform the background of the LCA in this study. (Details of the inputs and outputs of the model are provided at http://models.pbl.nl/image/index.php/Framework_overview).

Scenario Evaluation: The Wurst Software

IMAGE scenarios serve as a source of information to adapt the LCI background data (figure 1). Apart from being the most comprehensive and widespread LCI database, the ecoinvent database also has the advantage of distinguishing between two types of processes: transformation activities and markets (consumption mixes) (Wernet et al. 2016). This is an important feature because it simplifies identifying and changing parameters in ecoinvent when using IMAGE scenarios. To systematically approach the identification and changing of parameters in ecoinvent, we developed Wurst, a Python-based software that enables the systematic import, filtering, and modification of LCI databases. The current version of Wurst (available for download at https://github.com/IndEcol/wurst) focuses on ecoinvent and includes IMAGE scenario data as well as other sources. Other LCI and scenario databases are to be incorporated in the future. For this study, a specific functionality of the software was developed to link data formats of ecoinvent version 3.3 and IMAGE. The corresponding functions for import, filtering, and modification of LCI databases are provided in the supporting information available on the Journal’s website. For example, functions related to the regional match between databases in Wurst are used to generate ecoinvent LCI databases for different years into the future based on the IMAGE scenarios.

Data Import

We first imported ecoinvent and IMAGE scenarios data into Wurst, for which we wrote specific importing and cleaning functions. In particular, the “cutoff system model” of the ecoinvent database was imported (see Weidema and colleagues [2013] for details of this model). This means that monofunctional processes were adapted using the IMAGE scenario data to generate modified (future) monofunctional processes. After importing the data, we mapped the available technologies for both datasets (Appendix I in the Word file of the supporting information on the Web) as well as for all regions (Appendix II in the Word file of the supporting information on the Web). For the technology mapping, we assigned several related technologies in ecoinvent to an overarching IMAGE technology (Appendix I in the Word file of the supporting information on the Web) because ecoinvent provides more granular descriptions of technologies than IMAGE. Data for the overarching technologies in IMAGE are used to change the more detailed ecoinvent processes. Moreover, electricity generation technologies that will be relevant in the future according to the IMAGE scenarios but that are missing in ecoinvent were added to the latter to create an extended ecoinvent. These technologies are concentrated solar power (CSP) and carbon capture and storage (CCS), which we included using datasets from ecoinvent version 3.4 and from work by Volkart and colleagues (2013), respectively. For other technologies, such as natural gas combined heat and power generation with carbon capture and storage, which are missing in ecoinvent but less relevant in the future, we used proxy inventories from already existent technologies in ecoinvent (for all proxy technologies see Appendix I in the Word file of the supporting information on the Web). Technologies were left unchanged if they were related to other sectors, such as fossil-fuel and biofuel production, transport, and raw materials production. This choice is related to the focus of this study as a proof-of-concept as well as to the specific case study for which the electricity sector is most relevant, and it is not dictated by the IMAGE scenarios, which do include other sectors. In the discussion section, we elaborate on the possible implications of expanding the approach to other sectors, part of the IMAGE scenarios.

For the regional mapping, a one-to-one correspondence was assigned between IMAGE and ecoinvent regions where possible (Appendix II in the Word file of the supporting information on the Web). For regions in ecoinvent that involve more than one region from IMAGE, we used an average of IMAGE data. For smaller regions in ecoinvent, for instance, provinces in a country, we used the data of the larger region from IMAGE. An example of region and technology mapping is shown in figure 2, which illustrates that the electricity mix in ecoinvent has a closer match with that of IMAGE Western Europe, as electricity demand is dominated by Western European countries. In the interest of transparency, the complete region and technology mapping and the associated Python scripts are presented in Appendixes I and II in the Word file of the supporting information on the Web.

Figure 2. The 2012 electricity mix for Western and Central Europe regions in IMAGE and for the ecoinvent version 3.3 process Electricity, High-Voltage, Production Mix for the European Network of Transmission Systems Operators for Electricity (ENTSO-E). Technologies in ecoinvent are aggregated according to the map in Appendix I in the Word file of the supporting information on the Web and exclude the proxies for biomass steam turbine, oil combined cycle, and biomass combined cycle to show the original ecoinvent data without modifications.

Parameter Identification (Data Filtering)

Parameters from ecoinvent that are to be modified were identified according to the process name and unit of the reference output flow. For instance, for electricity production technologies that use coal, the ecoinvent process names include the words hard coal or lignite, and the unit of the reference output-flow is kilowatt hours (kWh). For electricity markets, the same reference output-flow unit is used, but the names include “market for electricity, high/medium/low voltage.” Such keys determine the processes that contain the parameters to be modified. These are technology-related parameters, that is, economic and environmental flows (input and outputs) such as greenhouse gas (GHG) emissions, for instance, carbon dioxide (CO2) emissions to air, or market-related parameters, that is, electricity market mixes in ecoinvent, such as technology shares in high-voltage electricity markets. Because the changes to ecoinvent parameters depend on the region and the technology, the corresponding IMAGE parameters were filtered from the set of total IMAGE output variables using the following filtering criteria: the years, the sector (in this case, electricity production), the overarching technology (e.g., coal steam turbine), the regions, and the scenarios of interest. This procedure generates two subsets of data, one from ecoinvent and one from IMAGE, which are related to one another via the region and the technology, as was explained in the previous section.

Parameter Changes

Starting with the ecoinvent and IMAGE subsets, we modified the ecoinvent parameters according to a number of rules (figure 3). For GHG emissions available in both ecoinvent and IMAGE (i.e., methane, sulfur dioxide, carbon monoxide, nitrogen oxides, nitric oxide emissions to air), we used the emission factors from the IMAGE scenarios as technology parameters, replacing those of ecoinvent for the different technologies. Using the IMAGE emission factors ensures coherency between the data used to describe the present and the future emissions. Differences between the emission factors in IMAGE and ecoinvent may be due to the use of different data sources and different methods to derive them. Most IMAGE emission factors are derived from the Electronic Data Gathering, Analysis, and Retrieval system (EDGAR) database (http://edgar.jrc.ec.europa.eu/overview.php?v=431), with emissions and activity data per sector and country, while ecoinvent uses mostly bottom-up or parameterized data per technology; for example, the CO2 emissions from burning coal in ecoinvent depend on the mass and carbon content of the coal burned in the process (Weidema et al. 2013). Emission factors in IMAGE were adapted by dividing them by the efficiency per technology in IMAGE because in IMAGE they are reported per MJinput and not per MJelectricity-output as in ecoinvent. All other flows (economic and environmental), for example, emissions other than GHGs emitted to air, were scaled using future technology efficiencies of the IMAGE scenarios for year i and scenario j. The final amounts of these flows, in their original ecoinvent units, were multiplied by a scaling factor (SF) calculated as shown in equation 1.

Figure 3. Schematic representation of technology and market parameters and their changes. Technology changes are presented in bold italics, and market changes are presented in underlined italics. Both are year dependent (i) and scenario dependent (j), as IMAGE data are year and scenario dependent. The scaling factor (SFi,j) was calculated as shown in equation 1

Further changes of market shares of electricity technologies are applied to high-voltage electricity markets in ecoinvent (Treyer and Bauer 2016). We replaced the shares of electricity-producing technologies defined in ecoinvent by the electricity mixes from the IMAGE scenarios. A different procedure was used for solar photovoltaics and small combined heat and power plants that supply electricity at the low- or medium-voltage level. We connected these technologies to the high-voltage level and assumed that all electricity generation is supplied at the high-voltage level. This procedure was chosen in favor of the systematic approach we propose, despite the error that this assumption might introduce, which we believe is small.1 Moreover, as only electricity markets change, transmission grid markets and SF6 emissions generated during transmission were not adapted and were kept at the original ecoinvent levels. These market changes are expected to capture system changes that are not necessarily related to technology efficiencies.

In the supporting information on the Web (Excel files), we present per-year tables, generated in the modification functions provided in the supporting information on the Web, with the changes made to technology and market parameters for one of the scenarios used in this study. The final output consists of future ecoinvent databases that are year and scenario dependent.

Life Cycle Inventory Calculation

The final step of the scenario evaluation involves the calculation of the LCI and characterized LCA results using the modeled future ecoinvent databases. Brightway2 (Mutel 2017) was used for this purpose. Brightway2 uses as input the future ecoinvent databases and calculates the inventory for the specified EV and ICEV (see case study section). The base year is 2012 because ecoinvent mostly represents the economy of this year. Selected future years are 2020, 2030, 2040, and 2050.

When the Background Matters: Using Scenarios from Integrated Assessment Models in Prospective Life Cycle Assessment

Table of Contents