Life Cycle Inventory Availability: Status and Prospects for Leveraging New Technologies

Abstract

The demand for life cycle assessments (LCA) is growing rapidly, which leads to an increasing demand of life cycle inventory (LCI) data. While the LCA community has made significant progress in developing LCI databases for diverse applications, challenges still need to be addressed. This perspective summarizes the current data gaps, transparency, and uncertainty aspects of existing LCI databases. Additionally, we survey and discuss novel techniques for LCI data generation, dissemination, and validation. We propose key future directions for LCI development efforts to address these challenges, including leveraging scientific and technical advances such as the Internet of Things (IoT), machine learning, and blockchain/cloud platforms. Adopting these advanced technologies can significantly improve the quality and accessibility of LCI data, thereby facilitating more accurate and reliable LCA studies.

Graphical Abstract

INTRODUCTION

Since its beginning in the late 1960s, life cycle assessment (LCA) has developed significantly to become the methodology for holistically evaluating the environmental impacts of products, processes, or services. An overview of the historical development is outside the scope of this article and can be found elsewhere (e.g.,¹). The development of the LCA ISO standards 14040 and 14044 (2006),² currently the generally accepted methodology, are key milestones in the adoption of LCA. The inclusion of LCA in legislation and regulations, for example, in the European Union Renewable Energy Directive,³ is also evidence of the importance given to LCA. Moreover, the growth of LCA in industrial assessments can be gauged from the exponential growth in businesses reporting on their environmental impacts through voluntary reporting initiatives such as the Carbon Disclosure Project.⁴ Other industry-specific initiatives, such as the Higg Index by the Sustainable Apparel Coalition, which uses LCA to evaluate the environmental performance of apparel and footwear companies and their products, have significantly increased the adoption of LCA in businesses. Nonprofit organizations like the International Society for Industrial Ecology, Society of Environmental Toxicology and Chemistry (SETAC), and the American Center for Life Cycle Assessment (ACLCA) have supported the LCA community. The United Nations Environment Program has also contributed to the development and promotion of LCA via the Life Cycle Initiative.⁵ From a scientific and technical point of view, publications increased at a 13% annual rate between 2016 and 2020 to over 6000 publications per year (see Figure 1 [search keywords: life cycle assessment analysis]).⁶ The growth of LCA encompasses many areas, ranging from health studies to environmental sciences. The range of applications requires the continued evaluation of LCA methodologies and resources. Life Cycle Thinking (LCT), for example, is a paradigm central to developing more sustainable processes and products and is one of the key principles of understanding the role of LCA methods, data, interpretation techniques, and other LCA components.⁷ LCA has demonstrated the capability to improve supply chain efficiency across various metrics. For example, energy and fossil fuel use are common LCA metrics that can improve the economic and environmental performance of a given product. The electronics sector relies on expensive and scarce materials like rare earth elements. LCA studies enable consumer electronic companies to efficiently track the provenance of their materials and identify opportunities to minimize environmental impacts across global supply chains.

Figure 1.

Growth in life cycle assessment publications (1980–present) [Source: Web of Science].

Life cycle inventory (LCI) data is a key component that underlies almost every aspect of LCA. LCI analysis is one of the four steps in the framework of LCA as defined by ISO 14040,² normally considered as the second step (after definition of goal and scope) in most studies. An LCI analysis broadly involves data collection, calculations, and allocation of flows and releases. Data collection can be a resource-intensive process. It requires an accounting of raw material inputs, energy used, other ancillary inputs, products, waste, water discharges, and emissions across a system boundary. The accuracy of life cycle impact assessment results hinges on the availability of high-quality LCI data. Hence, it is fundamental to have adequate inventory data to ensure that the calculated resource use and impacts are accurate and realistic, leading to appropriate conclusions and future actions based on the study results.

The increasingly widespread application of the LCA methodology has driven the development of LCI. Industrial and commercial institutions have led the way in gathering, sharing, and validating LCI data. Environmental compliance, consumer preference, and supply chain efficiency were key reasons for adopting LCA. The importance of these drivers varies by industry. Environmental compliance is a key factor in the building and construction industry and the chemical sector. Whereas, consumer preferences drive the energy and consumer goods sectors. Consumers are willing to prioritize clean fuels and products with low environmental impacts. Evidence that consumers would pay a premium for “green” products remains challenging. However, companies across various industries have made green pledges and other efforts to decarbonize their supply chains. This has led to growing efforts in scaling LCA data generation across entire product portfolios.

To ensure objective and robust analysis, as per the requirements of the ISO 14044 standard,² the use of primary data, which is data obtained directly from individual processes (for example, using sensors or other measuring devices inserted in the equipment), is preferred as it more accurately represents reality. However, obtaining primary data for vast complex product value chains is often challenging. Without primary data, other data derived from design-based calculations, process simulation, or market-based statistics or estimations can be used as secondary data. Several commercial and nonprofit organizations have compiled life cycle inventories based on primary and secondary data in LCI databases that are currently used extensively in LCA practice.

Researchers need to understand the trade-offs and gaps of various databases to conduct effective LCA. Even with so many LCI databases, the data availability remains a fundamental limitation that hinders the applicability of LCA in many research fields. We herein provide our perspective on addressing the challenges of existing LCI databases through the opportunities created by recent scientific, technical, and social advances.

This article outlines the current state and future direction of life cycle assessment inventory data. Its specific goals are to identify key challenges in LCA inventory management, propose innovative solutions to enhance data quality and accessibility, and explore the potential of machine learning and other computational methods in streamlining LCA processes. By addressing these aspects, the article aims to contribute significantly to the field of LCA and highlight opportunities for advancing the science and practice of the discipline.

The goal of this study is not to prescribe an overarching solution to the growth of LCI databases. Rather, we seek to draw awareness in hopes that the community identifies sustainable pathways enabled by emerging technologies to LCI interoperability, which is needed to reconcile LCA efforts across projects and research domains.

LIFE-CYCLE INVENTORY DATA

Life cycle inventory (LCI) data is growing rapidly as evidenced by emerging public and commercial databases that cover various applications, methodologies, and niche requirements.

Hence, many LCI databases were created in the past decades, making it harder sometimes to follow the evolution of the field. And more LCI data is being constantly created, as more and more products, processes and services are being studied. Even though it is not required to release the data to the public, there are currently efforts to promote transparency in life cycle inventory data developed by researchers and organizations. For example, current requirements in European Union and US funded projects require that data gathered in the project will be made freely available in open repositories, such as Zenodo or Data.gov. To assist LCA practitioners, searchable online aggregators have been created that group existing data sets and databases, of which Open LCA Nexus⁸ and the Global LCA Data Access (GLAD) network from the UNEP Life Cycle Initiative⁹ are the more comprehensive. Both services are being constantly updated, as more data or databases become available, and the searching facilities allow the identification of data sets for specific products or process units, facilitating the task of performing an LCI. However, their applicability remains limited, as they focus their attention on data and databases from companies and organizations that provide life cycle data sets, leaving out general purpose repositories such as Zenodo.

To understand the gaps in LCI data, we reviewed the databases listed on the OpenLCA nexus⁸ and other known resources. More than 27 LCI databases were identified and critically reviewed (see the Supporting Information and DOI 10.5281/zenodo.12385356), and a sample is shown in Table 1. The Supporting Information includes a summary of the LCI databases, their data gaps and limitations, their licensing requirement (free or open versus commercial), data format, and documentation availability. The number of LCI databases will change as more data becomes available, and databases consolidate, evolve either by including more data or by updating existing data for processes units/systems, or become obsolete to fulfill the needs of practitioners in various areas.

Table 1.

Sample of LCI Databases and Data Gaps or Limitations

Databases	Summary	Data Gaps/Limitations
Agri-footprint	5000 data sets on food, feed, and ag. intermediates.	Uses variety of databases for some background data.
Agribalyse	2500 data sets on ag. and food products produced/consumed.	France
ARVI	For circular economy. Developed between 2014 and 2016 as a part of a research program sponsored by CLIC Innovation Ltd.	Finland. Data not updated and relies on older versions of ecoinvent for background.
bioenergiedat	Developed in 2013 for biofuels, includes wood and wastewood, wheat, and biowaste.	Germany. Data appears not to be updated and relies on older background.
Carbon Minds	1300 data sets for chemicals and plastics.	Data appears to use ecoinvent data for background.

LCI databases vary in scope, regionality, and accessibility. Some databases encompass life cycle data for diverse applications and industries ranging from pharmaceuticals to fuels.¹⁰ Large LCI databases provide convenient access to comprehensive data sets. They also facilitate the use of multiple impact assessment methodologies. LCI databases have broadened their scope as the field has expanded the extent of analyses and indicators.¹ Large LCI databases require significant investment and support or a dedicated community of contributors to provide, curate, and validate data (e.g.,¹¹).

A growing number of LCI databases specialize in a narrower scope (see SI). For example, Carbon Minds (www.carbon-minds.com) focuses on chemicals and plastics. Worldsteel covers several steel products. The Agricultural Life Cycle Inventory (Agri-footprint)¹² provides detailed information on agricultural products and processes, and NIST (National Institute of Standards and Technology) Engineering Laboratory developed the Building for Environmental and Economic Sustainability (BEES) software with data for assessing building products and construction materials.

Regional databases that gather data by region, such as the European Union or a country, are another type of focused LCI database. The region can be specified based on resource provenance or end-use market. The Australian National Life Cycle Inventory Database (AusLCI)¹² covers Australian LCI data for various sectors. The Norwegian Petroleum Directorate’s LCA¹³ database focuses on Norway’s oil and gas production. For international supply chains, it becomes increasingly important to assign life cycle inventory appropriately. For example, the manufacture of electronic devices involves gathering resources from different mining sources, shipping the resources to manufacturing facilities, and delivering the final products to end markets. The importance of environmental impacts will depend on the region where the activity and its releases occur. For example, ecotoxicity would be most critical in regions with species facing extinction.

Finally, accessibility encompasses the licensing requirement, data format, documentation, and other factors that facilitate a broad audience’s use of the LCI database. Free or open LCI databases allow broad community use and even participation in developing LCI data. Commercial databases trade openness for institutional or financial support. Hybrid approaches are also available where a commercial LCI database can solicit public input or a free LCI database receives corporate support. Due to the variety of data formats and methodological choices, the interoperability of LCI data across different platforms is a challenge. Some of these data formats are exclusive to a particular database of LCA software. Efforts such as the International Life Cycle Data (ILCD) system data format developed by the European Commission’s Joint Research Center (JRC) are steps in the right direction. ILCD uses an open-source Extensible Markup Language (XML) data format to facilitate collaboration across different software platforms.¹⁴ However, challenges such as variability in the elementary flow list still exist, which can be addressed by implementing harmonization to a common elementary flow list. A common data format does not guarantee compatibility. JSON (JavaScript Object Notation) and XML, for example, are open formats that any plaintext editor can read or write, but the structure of the data determines whether it is compatible with LCA software.¹⁵

DATA STRUCTURES AND DATA GAPS

LCI databases organize data using data structures.¹⁶ Figure 2 shows an example of interconnected LCI data structures. In this example, an LCI project contains multiple product systems, each with multiple processes. Processes connect multiple actors, sources, and flows. LCA methods are often applied at the flow level by multiplying impact characterization factors with elementary flows (i.e., exchanges with the environment) like material or energy data organized by unit groups. Data structures mirror LCA standards by defining the connections between flows and processes. These include connections to product systems, life cycle impact assessment (LCIA) methods, flow properties, and others. The underlying concepts of most LCI data structures translate across various commercial and free LCI databases. However, limited interoperability often makes sharing data between LCA applications difficult, thus contributing to data gaps.

Figure 2.

Life cycle inventory database data structures, with one-to-many relationships between the principal elements (adapted from ref 17).

The size and scope of existing LCI databases are large. Ecoinvent, for example, has over 20,000 data sets encompassing material and energy flows, unit operations, systems, and others. However, LCI databases remain small relative to the potential number of systems and processes in operation globally. Fields like energy, consumer appliances, and large-scale industrial applications are heavily documented. LCA standards and their application are well-suited to evaluate commercially mature technologies.^18–20 Data gaps remain despite growing efforts from the LCA community.²¹ Data gaps exist because the pace of technology development exceeds technology assessment, some disciplines lack the knowledge or tools to gather, interpret, or share LCI, and some scientific or technical advances are needed to collect specific LCI data.

LCI data is often gathered as part of environmental compliance efforts, but a growing number of voluntary efforts exist. Private companies, including environmental consultancies, often collect data to demonstrate compliance with local and national regulations. Open science and Open Data requirements linked to scientific and innovation projects, as is currently the situation in Europe, also lead to the release of LCI data. This LCI data is abundant, but unfortunately, it is usually proprietary or not packaged for public consumption. An additional legal, technical, or economic effort is required for this information to be translated into useful LCI. Nonprofits, open-source, and other community efforts are finding new ways to support translating this information into public LCI databases due to growing environmental impact awareness. Both private and public efforts would benefit from accessible LCA software that enables rapid assessment of new technologies.

The scientific community is embracing LCA to assess increasing technological and scientific advances. Most of this effort is voluntary and unsupported. While LCA has been around since the 1960s, LCI data gaps are significant. More than 200 million organic and inorganic substances have been disclosed in the literature since the early 19th century (https://www.cas.org/about/cas-content), but LCI data is available for only a very small part of them. LCI databases include at most a few thousand chemical compounds. This situation occurs for many chemicals used in industrial practice. For example, the textile sector employs more than 15,000 chemicals, of which around 10,000 are dyes, with the overwhelming majority of them lacking proper LCI data.²² A simple explanation for this gap is that most scientists/specialists are not trained or do not have the resources to conduct LCA. Moreover, questions of intellectual property or private/confidential data may play a significant role. Companies may be unable or unwilling to release information about their processes, including emissions, due to competitiveness or legal considerations.²³ Incentives (e.g., Data Openness Badges by Journal of Industrial Ecology) could encourage practitioners in many scientific and technical/industrial fields to gather and share more LCI data. This would allow the development of more rigorous and objective LCA studies. This would allow the development of more rigorous and objective LCA studies.

Some practitioners face unique challenges in gathering LCI data. They may lack adequate knowledge of the process systems necessary to assess all of the data needs and how it can be obtained. For example, process systems may involve diverse reactions and physical processes between various chemicals and materials. These compounds are often quantified, but the nature of their formation, decomposition, and other characteristics are difficult to measure reliably, and that information is crucial to properly evaluate the environmental impacts associated with their production. These measurements improve as analytical equipment and techniques develop. Once measured, LCA practitioners must decide the scope and type of LCI data to gather and its storage format. After all of the effort to measure, curate, and store the data, the decision has to be made to share the information with the community. Journal publications are the most common format, but the odds that LCI data from a journal publication contributes to established LCI databases are low. The process for submitting and validating user-supplied LCI data varies by database, which presents an additional barrier. Some frameworks, like GLAD, and software developers and suppliers are currently working to lower these barriers.

MEASURED AND INFERRED DATA GENERATION

In conducting a life cycle assessment (LCA), it is crucial to outline the product or service system’s supply chain. Static and aggregated depictions of supply chain relationships are apt for stable systems but are not sufficient for new and fluid ones.²⁴ The use of big data, for instance from social media, can introduce a socio-economic perspective to LCA. Moreover, big data that captures the nuances of human movement dynamics, like geotagged data from social networks, can significantly refine LCA of transport systems by factoring in the variations in travel behavior (e.g.,¹⁹). Such data is instrumental in promoting the evolution of advanced transportation technologies, including electric vehicles. Cai and Xu,²⁵ for example, leveraged the travel patterns of over 10,000 taxis to study the effects of electric vehicle design, economics, and government subsidies on electric vehicle adoption, usage, and the associated greenhouse gas emissions.

Incorporating measured data, such as the environmental release top-down databases as described in Meyer et al.²⁶ and the bottom-up process specification methods in Smith et al.,²⁷ into LCI databases is a well-established resource. However, inferred data is a promising technique that can also be used to gather data. There are various forms of inferred data, and sensors can contribute to both measured and inferred data. For example, building energy consumption can be inferred from ambient temperature measurements and known facility characteristics. Physical inventory movements can be derived from business transaction data. The convergence of Internet of Things (IoT) and low-cost, high-performance computing provides new possibilities for real-time assessment of LCI data.²⁸ The next step involves incorporating feedback loops that adjust system operating conditions based on LCA data.²⁹ Industry 4.0 (I4.0), also known as the Industrial Internet of Things, refers to the next stage of evolution in the manufacturing sector that takes advantage of advances in Information and Communication Technology, automation, machine learning, and real-time data to better understand and control the various supply chain and manufacturing phases. I4.0 uses smart sensors, real-time data processing, and interconnectivity to drive efficiency, predictive maintenance, enhanced field service, energy management, and asset tracking.

The building and construction sector is one example where these technologies have been successfully implemented. Construction companies must ensure their materials do not release toxins and other harmful compounds. Thus, they track most materials’ provenance, manufacture, and deployment. Construction companies are adopting novel sensors that track environmental contaminants (with, for example, air sensors³⁰) throughout the lifetime of the building. Sensor measurements are often exclusive to the construction company, but decreasing costs and improved information technologies like the IoT enable consumers to track sensor measurements. The rapid growth of IoT-capable sensors would increase the available data to augment building and other LCI databases.³¹

MODEL-BASED AND STREAMLINED INVENTORY

When models that give an adequate system description are available or can be developed, they may be used to obtain inventory data. Depending on the underlying principles or modeling assumptions, a spectrum of LCI-based models may be defined. These models range from simpler empirical or causal models based on the interrelations between the various process systems to models built from rigorous physical and chemical principles. Empirical methods were used in the initial years of LCA methodology development^32,33 for inventory adjustment when a process is scaled-up using matrix methodologies.^32,34

An interesting model-based approach takes into account the network nature of LCA, in particular the upstream processes needed to obtain the materials, energy, and other services required to obtain a given product or service. In general, from a chemical and practical point of view, to obtain a given compound one needs to use several chemicals and perform certain activities, that may be chemical reactions and/or separation process. Hence, when the LCI of a certain compound is not available in the literature or in existing LCI databases, one may follow the upstream steps needed to obtain it, using available information on synthesis procedures and chemical processing, until it reaches a level in which the LCI of all compounds is known. In practice, a life cycle tree of the compound is created with the upstream processes needed to obtain it. Then, an approximate LCI of the unknown can be estimated using the known LCI and the stoichiometric and chemical information.

When compared to other approaches, for example using LCI of similar compounds, the life cycle tree strategy tries to include as much information as possible from the production process, ensuring a more realistic estimate. When a direct simulation of the production process is too complex, or no sufficient computational resources are available, this is a valuable and relatively easy approach. Even though it has advantages, the life cycle tree approach is not commonly used in practice because of logistic challenges, confidentiality constraints, and other data-gathering limitations, as for example difficulties in obtaining thermodynamic and/or conversion data. Examples of application of this methodology include the work of Zhang et al.³⁵ and Cuellar-Franca et al.³⁶ that used it to obtain estimates of the LCI of ionic liquids, and Carneiro et al.³⁷ Attempts to develop general methodologies for the estimation of LCI data based on the life cycle tree method were proposed by some authors.^38,39 The existence of chemical processes, compound properties, and emission databases that aggregate specific or average data open up the possibilities of using this methodology to evaluate missing LCI data using data mining tools, as already shown by Cashman et al.⁴⁰

The increased availability of computational power and software with more advanced capabilities (e.g., better physical property prediction) has led to the utilization of more rigorous models for inventory prediction. Among the various possibilities, chemical process simulators are seen as a viable option and have been used extensively in practice.⁴¹ Their utilization is particularly valuable when designing a production process for which some or even all data is not available or is unreliable. Moreover, process simulators can also be used to estimate inventory for components and/or process units when they do not exist at commercial scale. Some examples of the application of process simulators include Parvatker and Eckelman⁴² who estimated LCI data for 151 chemicals through streamlined simulation models, the work of Mata et al.,^43,44 which analyzed the impacts of gasoline blending components, Smith et al.⁴⁵ that studied how process simulation can be used to design more environmentally conscious chemical processes taking into account fugitive and open emissions, and Azam et al.⁴⁶ that analyzed and compared different options for plastic waste thermal treatment processes.

In other words, various simulation tools were combined together, each one aimed to address specific physicochemical phenomena or other aspects that are relevant to the system under analysis. An interesting example is the work of Arora et al.⁴⁷ that combined Computational Fluid Dynamics (CFD) software with a chemical process simulator in the analysis of ammonia production using biomass as a hydrogen source.

Despite the potential of the previous approaches, there are systems for which rigorous models are not available, or are too complex, or the required data and/or resources are not available, precluding the evaluation of the LCI needed to perform an LCA study. These situations are particularly relevant in complex systems that involve many different parts, components, and even involve different stakeholders; the latter being significant in particular for global supply chains that are common nowadays. A potential solution considers the utilization of hybrid models that combine first-principles with data-based models, such as neural networks or machine learning. The biotechnological or pharmaceutical processes, where it is still difficult to have adequate models that adequately describe their behaviors, are where hybrid models are seen as having much potential, as they combine mechanistic and data-driven models to minimize the problems of each one.⁴⁸

The availability of data and information and/or adequate models that describe the various system parts can be used in some situations to simplify or streamline the development of the LCI for a given process system. When doing this, attention must be given to the potential loss of accuracy when creating the inventory, as the streamlining process itself may oversimplify data acquisition and processing. Nevertheless, the reduction in complexity coupled with faster and simpler analysis may justify streamlining the inventory and eventually implementing these computational methods.

An interesting example of LCI streamlining is the work of Sacchi et al.,⁴⁹ which coupled Integrated Assessment Models (IAM) to obtain LCI to perform prospective LCA studies. This is a particularly complex area as it is necessary to predict the evolution of existing and future conditions, for technologies, socioeconomics, and targets for the environment, and the LCI should reflect these issues. The authors have developed an open-source tool implementing their methodology, PREMISE (https://github.com/polca/premise), and applied it to a set of case studies that included activities with an actual or future impact in sustainability, such as direct CO₂ capture from air.

Another possibility for streamlining LCI involves using standard environmental information datasheets, particularly Environmental Product Declarations (EPD) or International Material Data System (IMDS). In some areas, for example, the construction or the automotive sector, the information available may be enough to make an LCI. Data mining and/or text extraction may play an important role, as the EPD knowledge database is increasing fast due to regulatory and consumer pressure. De Oliveira et al.⁵⁰ considered the utilization of supplier IMDS to create inventories in the automotive sector and concluded that particular attention should be given to the cutoff criteria, and what materials are accounted for in the inventory. New technologies based on artificial intelligence or machine learning can help identify the best data-driven models according to the process and/or data characteristics.

MACHINE LEARNING AND MODEL DATA GENERATION

Machine learning and model data generation are emerging techniques with the potential to enhance the scope of LCI databases. Machine learning can bridge the gap between various uncertain data sources and LCI databases by interpreting, validating, and generating LCI data. Inferred data is often employed when measured data is unavailable. For example, Turner et al.⁵¹ employed satellite and surface observation data to infer methane emissions. This data carries uncertainty. A machine learning model can combine inferred data with domain knowledge-generated data and information from other sources to reduce the estimate’s uncertainty. This feedback loop could be leveraged to decrease the uncertainty of life cycle inventory databases. Verifiable data from IoT and Blockchain networks can be introduced to validate and enhance this feedback loop. When connected to IoT and Blockchain networks, machine learning algorithms could develop adaptable and scalable capabilities that tackle large-scale sustainability assessment needs. For example, sensors located in waterways can be connected to have real-time measurements of water quality and couple the data with information on fertilizer use. The sensors provide a basis for measured data, and the fertilizer use would be gathered from generated and inferred data stored in blockchain networks.

Machine learning techniques encompass various methodologies like natural language processing (NLP) and artificial neural networks (ANN). NLP has been applied to automate the information extraction for chemical and material syntheses.^52,53 The general workflow of such NLP-based pipelines starts with tokenizing sentences, creating a “bag-of-words” vector, and tagging the part-of-speech (POS). The relevance of the paragraph/sentence (e.g., “Method” section) can be determined by classification models (e.g., logistic regression classifier) that are trained on the manually labeled samples. Heuristic rules (e.g., matching keywords to chemical names of interest) and machine learning models (e.g., ANN) can be combined to extract the synthesis information, such as the number of reactants, reaction temperature, yield, etc. Jensen et al.⁵⁴ connected an automated information extraction pipeline with a random forest model to predict the germanium zeolite framework density from the synthesis conditions extracted. Predictive synthesis using NLP-based information extraction pipelines has the potential to identify the trend of technology development (e.g., a trend of increase in reaction yield as a result of changing reaction conditions) and, hence, may be used to generate future foreground inventory data representing plausible technology advancement.

Song et al. investigated the use of machine learning to address data deficiencies in LCA.⁵⁵ Their work focused on using ANN for rapid life cycle impact screening, deriving chemical sensitivity distributions, and reducing the uncertainty of characterization factors. The machine learning model was trained using molecular structure information to estimate six impact categories. The ANN achieved R² values of over 0.71 in estimating acidification, human health, and an overall LCIA method, eco-indicator 99. In their second study, Song et al. investigated the use of machine learning to estimate the lethal concentration of chemicals in aquatic species based on molecular structure. The average R² for this analysis was 0.67. Gao et al. developed a framework for machine learning models for facility life cycle cost analysis.⁵⁶ The motivations for this work were to “address the shortage of life-cycle cost data and the complexity of predicting real future costs.” Life cycle costs are closely tied to LCI through common technoeconomic analyses.

Sousa and Wallace evaluated the use of machine learning to support product classification in the approximate life cycle assessment of design concepts.⁵⁷ The idea was to train ANNs using detailed descriptions and LCA data to generate simplified inventory data based on concept descriptors of new products. These descriptors included composition, expected lifetime, additional consumables, and others. Examples of classifications included determining the products’ feasible end-of-life strategies; classification criteria like aspects of product use that have binary answers (i.e., yes or no); and product categories, for example, stationary products with internal energy consumption in use. The surrogate LCA machine learning models had an observed error rate of 16 to 21% in classifying different product descriptors.

D’Amico et al. published a call for data to help with machine learning models for building LCA.⁵⁸ The aim was to, for example, develop a machine learning model that can predict building LCA features (a form of LCI) based on a visual representation of building structures. Their effort was built upon an online submission platform called the Resource Efficient Built Environmental Lab (REBEL), where participants can upload structural and other attributes of buildings. The group planned to conduct LCA on all submissions to develop training data for the machine-learning models.

Zhu et al. recently published an application of LCA and machine learning for high-throughput screening of green chemicals.⁵⁹ This study employed molecular descriptors to develop predictions of impact factor values based on previous LCA of known chemicals.

Previous studies can be categorized into two types of machine learning applications based on their outputs: direct prediction of LCI (e.g., energy consumption or emissions) or life cycle impact assessment (LCIA) results (e.g., the end point or middle point impact using different LCIA methods). The methods for LCI predictions are likely to have more versatile applications for LCA practitioners, given the ability to choose different LCIA methods and customize LCI data sets for different projects. LCIA applications are likely more attractive for projects with screening purposes where an explicit collection of LCI data for individual materials/products is not critical.

Direct applications of machine learning without using knowledge-based models tend to rely on the metrics of similarity (e.g., similar chemical structure, process design, or data structure) between the investigated system and the systems with LCA data already available. This approach limits machine learning applications for emerging technologies that do not have sufficiently large data (LCI or LCIA) for similar processes. Therefore, some studies combined machine learning with knowledge-based, domain-driven models to generate the LCI data for new technologies or systems. Liao et al. used an ANN model to predict the yield of activated carbon production that was used as the input of a kinetic-based process simulation to generate LCI data of activated carbon made from diverse biomass species.⁶⁰ Cheng et al. used random forest (RF) to predict the yield and characteristics of biochar produced from slow pyrolysis, which was used in a mass and energy balance model to generate LCI data needed by the LCA.⁶¹

The use of domain-driven knowledge (DDK) models has the potential to transform the LCA discipline by accelerating the process of generating and validating LCI data. However, human confidence in DDK models may take some time because of the legal and ethical implications concerning the use of synthetic data. Nevertheless, this is a promising field for researchers to investigate as the applications of machine learning models develop.

BLOCKCHAIN AND CLOUD PLATFORMS

The availability of reliable supply chain data is key to high-quality LCA results. However, as the production and distribution of goods become more globalized, supply chain inventory management becomes increasingly complex. This is due to the challenges of tracking product origins, processing, and transportation, as well as the availability and uncertainty of data. Supply chain management faces both qualitative and quantitative challenges, and the main obstacle remains in creating a traceable and well-managed data system.⁶² However, blockchain technology provides a new solution to supply chain management challenges. By integrating blockchain into the supply chain architecture, a reliable, transparent, authentic, and secure system can be created.

Blockchain is a distributed ledger technology (DLT), which is a consecutive list of recorded data that is time-stamped and linked sequentially using cryptography, usually digital transactions.^62,63 The technology has four important characteristics: decentralization, persistence, anonymity, and auditability.⁶⁴ When business operations use blockchain technology as a distributed database, they can operate without relying on third parties. This means they are less vulnerable to malicious attacks, malfunction, and artificial alterations compared to traditional business operations that rely heavily on a centralized authority or third parties like banks.

In recent times, there has been a significant rise in the use of blockchain applications. These applications have expanded and are now being used in various areas, such as blockchain-enabled physical distribution and logistics, business process management, information sharing, business operations, and risk analysis.⁶² For example, to achieve decarbonization and alternative energy in the transportation sector, a thorough life cycle assessment (LCA) is required. One way to accomplish this is by utilizing blockchain technology, which can link multiple stakeholders and securely track data with its distributed architecture and immutable record characteristics.⁶⁵ This ensures that the data can be verified and trusted. Various industries have reported real-world applications of blockchain technology.⁶⁶ Blockchain technology has a significant role to play in LCA, particularly in generating LCI data and reducing uncertainty.

In a study conducted by Karaszewski and colleagues, they presented a blockchain-based LCA framework that integrates blockchain technology to ensure traceability and transparency in the goal and scope definition.⁶⁷ This framework also incorporates the Internet of Things (IoT) concept to collect and integrate real-time data at the inventory analysis level. Additionally, it creates an analytic form for impact assessment, as shown in Figure 3. The integration of blockchain and other smart enabling technologies, like sensors and devices that generate vast amounts of real-time data, can enhance the efficiency and effectiveness of LCA processes.⁶⁸

Figure 3.

Blockchain-based LCA framework showing the connections between traditional and blockchain-based LCA frameworks.

Examples of blockchain-related technical issues are throughput, security, scalability, and interoperability.^62,64 Specifically, the potential technical issues associated with the blockchain-based LCA framework are data manipulation challenges, big data storage challenges (scalability), and data transmission challenges.⁶⁸ A global push for decarbonization has seen value chain partners share product carbon footprint data. Blockchain technology can play a key role in enabling LCI/LCA data sharing among organizations securely and at scale.

TRANSPARENCY AND UNCERTAINTY

Transparency and uncertainty have become important in LCI database development. LCI transparency is defined as knowledge of the source, collection method, and output interpretation of the original data. Uncertainty types include epistemic uncertainty (i.e., reducible by acquiring a better understanding of the system) and aleatory uncertainty (i.e., irreducible due to the random nature of the system). The epistemic uncertainty of LCI data may stem from measurement errors or insufficient sample sizes, while aleatory uncertainty lies in the stochastic behavior of the systems (e.g., technological variations).⁶⁹ Transparency and uncertainty directly impact the quality of existing LCI databases.

The lack of transparency in LCI stems from the limited documentation that accompanies many databases.⁷⁰ These limitations range from lack of reference data to incomplete or incorrect documentation. Limited transparency is often accidental or due to limited resources, but it is sometimes intentional. Companies may not want to share the source of their LCI data to avoid legal exposure. Authors may not like to describe the collection method because it may not follow best practices. LCI databases are addressing these challenges by increasing the minimum quality required for data submission, which poses an additional barrier (although likely appropriate) to gathering LCI data and filling data gaps.

Uncertainty is inherent in almost every LCI data point, as discussed in another current perspective.⁷¹ The prevalence and contribution of different uncertainty types vary for each LCI data point. The complete removal of total uncertainty is impossible, but reducing uncertainty to a practical range is possible. Gathering LCI data is often expensive, which limits the effort to reduce uncertainty. For example, measuring LCI data for specialty chemicals can be challenging due to their limited quantity and variability along the supply chain and processing steps. Scientific advances in statistics and data science and technological advances in sensor and computational capabilities contribute to decreasing uncertainty in LCI data.⁷²

FUTURE DIRECTIONS OF LIFE-CYCLE INVENTORY DATA DEVELOPMENT

The future development of LCI databases involves the coalescence of scientific, technical, regulatory, and social advances. These advances will enable the collection, generation, and validation of LCI data from a growing community of practitioners.

The rapid growth of the LCA literature suggests a growing demand for collecting LCI data across scientific disciplines for increasing applications, whereas a combination of limited knowledge and resources results in incomplete or incorrect LCI data.

The LCA community is developing novel methods to generate verifiable LCI data. These methods leverage advances in information technology, computational resources, and machine-learning techniques. For better collaboration in the supply chain, industries are developing and consolidating common data structures that enable interoperability among LCI databases. Scientific and computational advances are enabling the measurement and generation of LCI data. Open data and cloud computing platforms are increasing the rate of data sharing and validation. Machine-learning models equipped with domain-specific knowledge, access to real-time and historical data, and broad communication capabilities could transform the generation and use of LCI data.

LCI data validation remains challenging due to limited transparency, inherent uncertainty, and legal and social barriers. Blockchain-based LCA frameworks could address these challenges by linking LCI data collection to community-based validation efforts.

The future of LCI data shows promising opportunities for scientists, engineers, and policymakers to develop methods, technologies, and protocols that accelerate the sustainability assessment of emerging products, systems, and resources. We encourage the LCA community to discuss these opportunities and develop roadmaps prioritizing much-needed advances in LCI databases.