Sustainability assessments are increasingly critical for evaluating the environmental impacts of business activities. Life-cycle assessment (LCA) is a key methodology in measuring these impacts, but integrating and analyzing data from diverse enterprise data sources to compute LCA indicators remains a challenging task. In this paper, we propose an approach that leverages Knowledge Graphs to formally model LCA indicators and their associated mathematical calculation formulas. The graph serves as a flexible schema supporting enterprises in documentation and mathematical interpretation of indicators for LCA, by linking their internal data sources to the Graph. On top of it, a suite of reasoning-based services is presented to automate the calculation of indicators from available data sources through algebraic manipulation, and to facilitate collaboration among a network of enterprises by enabling consistent comparison of sustainability assessments.
ceur-ws.org supply chain, it has to evaluate diferent impact indicators for each activity according to the impact category involved. This process requires information about products, emissions, pollution, waste on each activity, thus making the LCA a data-intensive procedure.
Many databases collecting LCI data are available with the aim of supporting sustainability assessment
within organizations. Among these, one of the most relevant is Ecoinvent2, which collects global data
containing detailed information on human activities and related environmental impacts. Ecoinvent
datasets also store mathematical relationships on diferent fields of an activity to calculate quantitative
measures, such as the amount of a certain emission. These formulas can serve as validation and support
for the calculation of certain measures, avoiding errors in the final assessment of impacts. In this
context, an important challenge concerns the lack of formalization of LCA concepts and standardization
across LCI databases. In fact, most LCI databases are based on traditional RDBMS systems, without an
explicit semantic support, thus rising interoperability and scalability issues [
In this paper, we address the mentioned issues by proposing a semantic-based approach to support accurate enterprise-tailored assessment of environmental impact. The approach is based on a Knowledge Graph for Ecoinvent data, focused on the explicit and formal representation of LCA indicators and their associated mathematical calculation formulas. The LCA KG is aimed to support an accurate calculation of environmental impacts, thus supporting its certification process by facilitating the documentation of the assessment and its sharing. The schema underlying the KG is based on several ontologies, including the Ecoinvent vocabulary and KPIOnto for representing indicators and related formulas. On top of the model, we propose reasoning-based services based on an algebra system to perform several supporting tasks. Primarily, it enables the automatic calculation of impact indicators from enterprise data sources through algebraic manipulation of formulas. Additional services allow the derivation of alternative calculation expressions for a given indicator and, in case of collaborative enterprises, the calculation of dependencies among indicators across diferent organizations to ensure consistent comparisons of sustainability assessments.
The remainder of this article is structured as follows: Section 2 discusses the design of the LCA Knowledge Graph, while Section 3 introduces logic-based services providing capabilities useful for formula calculation and comparison of impact results. Finally, Section 4 draws conclusions and outlines future work.
This section aims to discuss the design of the LCA Knowledge Graph, starting from the information stored in the Ecoinvent database and its Data Glossary representing its Linked Data representation. Then, we discuss the KPIOnto ontology for the representation of mathematical indicators and its use for the definition of the integrated schema for the Knowledge Graph.
The Ecoinvent database contains a large number of datasets, organized by multiple categories, such as sector (e.g., Chemicals, Electricity,), geographical area, and activity type (e.g., a transforming activity). The building blocks of the database are the so-called Unit Processes (UPR), also named activities. Each activity represents a single human process that can occur in a supply chain and is characterized by lfows, namely elementary exchanges (EEs) and intermediate exchanges (IEs). The first are exchanges from/to the environment and refer to the case in which an activity consumes natural resources or releases emissions. The latter are flows with other human activities and do not involve exchanges with the environment. IEs may output diferent products or wastes, including a reference product, that is the main driver of the activity. Exchanges have a set of properties, such as the water or carbon content that may serve for analysis e.g., the carbon footprint of the related product. All these information can support LCA with the evaluation of potential impacts to the environment of a specific activity. An interesting piece of information are the mathematical formulas associated with exchanges, properties or parameters. Parameters are specific type of values that can be used to express relationships, such as the eficiency, the correlation of input to variables and the estimation on emissions. Several quantitative measures can be expressed with mathematical relationships, including the amount of exchanges.
For example, let us consider the activity silicone product production3. Among its EEs, two relate to water emissions respectively to the air ( _ _ ) and to water ( _ _ ). Both EEs are associated with a mathematical expression calculating their amounts in the activity (expressed in 3), as shown below: 3https://ecoquery.ecoinvent.org/3.10/cutoff/dataset/6368/documentation ) (1 −fraction_TW_to_air)+
The tap_water_input refers to an IE which exchanges with another activity whose reference product is tap water. The water_cooling_UNO_input is an EE corresponding to the input of cooling water (where UNO stands for ‘unspecified natural origin’). The water_well_in_ground_input is another EE referring to the input of well water from the ground. The fraction_TW_to_air, fraction_CW_to_air and fraction_PW_to_air are parameters quantifying the amount of tap water, cooling water and process water emitted to the air, respectively. Another mathematical relationship expresses the equivalence between fraction_TW_to_air and fraction_PW_to_air. Also, fraction_CW_to_air can be calculated as (0.5 ∗ fraction_CW_OT_to_air) + (0.5 ∗fraction_CW_R_to_air), where fraction_CW_OT_to_air and fraction_CW_R_to_air are the amount of cooling water emitted to the air, respectively once-through system and recirculating system.
Besides being used to characterize how to calculate the amount for a flow, formulas can be expressed also to calculate overall impact indicators, which is essential to conduct LCIA of an activity or the entire supply chain. In particular, Ecoinvent provides documentation for the calculation of LCIA score of activities or individual exchanges, starting from the selection of an impact method. An impact assessment method can include diferent impact indicators, focusing on the evaluation of single footprints or impact categories. The LCIA score of an activity with reference to a selected impact category can be computed as: = ∑ , , where is the amount of the flow which impacts on , and is the coeficient. For instance, chosen the impact method EF, it is possible to assess the impact category ‘water use’ of the silicone product production activity by considering all the flows involved in water consumption. Thus, their amounts are multiplied with the corresponding coeficients to obtain the LCIA score which evaluates the environmental impact of the activity on water usage. Although Ecoinvent provides averaged amounts to calculate approximate LCIA scores, a more precise calculation, relying on actual enterprise data, is essential to precisely certify environmental impacts.
The Ecoinvent Data Glossary4 provides a reference for all metadata used within Ecoinvent inventory datasets, in the JSON-LD format, with the purpose to establish a common understanding of data. Its schema includes classes representing basic concepts such as elementary-exchange and intermediateexchange with their description, classification and formula, exchange-properties and parameter with the unit of measurement, and activity-name. At the time of the writing, with respect to the information available in the Ecoinvent datasets, the schema is incomplete and data is missing. For example, the relation between an activity and the exchanges is not defined, mathematical formulas are not reported for any flow, and activities’ properties are not detailed. Furthermore, the schema assumes formulas are encoded just as a string, namely a Text datatype from schema.org.
In order to explicitly represent the mathematical expressions to calculate amounts for Ecoinvent flows,
we resort on the KPIOnto ontology5. The ontology provides classes and properties for describing (Key)
Performance Indicators on the Semantic Web. The main class of the ontology is Indicator, which
describes the quantitative metrics enabling performance monitoring. It is described through a set of
properties including a description (KPIDescription), a unit of measurement (unitOfMeasure), the
aggregation function (aggrType), its business objective (hasBusObj), a set of dimensions of analysis.
An indicator can be either atomic or compound, built by combining several lower-level indicators.
Dependencies of compound indicators on their building elements are defined by means of algebraic
operations, that is a Formula capable of expressing the semantics of an indicator compositionally.
A formula describes how the indicator is computed and is characterized by the operator (property
hasFunction) and one or two operands (respectively, for unary and binary operators), which are
instances of FormulaArgument. In turn, each argument has a value, represented as an instance of
ArgumentValue. Finally, a value can either be a Constant, another indicator or, recursively, a formula.
The ontology has been used for various applications, ranging from formally representing a library of
KPIs [
The LCA Knowledge Graph has been defined by referring to a schema integrating the Ecoinvent vocabulary and the KPIOnto ontology. As shown in Figure 1, we introduce a property hasIndicator from a custom vocabulary (with namespace kg) associating each elementary/intermediate exchange and 4https://glossary.ecoinvent.org/ 5http://w3id.org/kpionto (a) (b) property with its amounts, represented as an instance of class Indicator. In this way, the information related to the calculation of flow amounts is represented using KPIOnto terminology. Furthermore, we introduce a relation kg:dependsOn between two indicators, which is defined through a SWRL rule, stating that the former indicator is compound and the latter is one of its components.
In Figure 2a an example of the resulting LCA Knowledge Graph is shown, representing, for lack of space, a subset of the indicators and formulas related to the elementary exchange water_to_air, for the silicone product production activity. The indicator ind_water_to_air is represented as a compound indicator with a formula defined as the application of an operand to a set of operators, which in turn are other indicators, e.g. ind_tap_water_input or ind_fraction_TW_to_air.
The model is designed to be general and versatile, serving multiple scenarios across diferent
organizations. In practical situations, an enterprise may have data available for only some of the indicators
in the model. In such cases, a mapping can be defined to explicitly state that a particular data source
contains data related to a specific indicator, following the model proposed in [
The LCA Knowledge Graph can enable an easier documentation of indicator definitions and LCIA results, ultimately making their understanding easier and shareable on the Web using open and a FAIR format. A set of services can be built on top of the LCA Knowledge Graph to support (1) quantitative impact assessments and (2) advanced analysis and comparison of LCIA.
A quantitative calculation of the impact is needed for more precise assessments and to validate the
accuracy and consistency of data against regulatory standard, reducing the risk of non-compliance
due to data errors. The calculation of indicators relies on an algebra reasoning system, derived from
PRESS [
When indicator definitions are mapped to the corresponding data sources at enterprise level, the formula manipulation functionality can be exploited to support the organization in calculating an indicator from the available data. To make an example, let us image an organization needs to calculate the indicator ind_water_to_air for the corresponding elementary exchange. Let us suppose the set of indicators available to the enterprise are the seven provided by the data sources in Figure 2b. As such, the formula for the required indicator cannot be directly computed, since there are three indicators which are not provided by any source, namely ind_fraction_TW_to_air, ind_fraction_PW_to_air, and ind_fraction_CW_to_air. However, by considering the indicators’ formulas, the algebra system derives that ind_fraction_CW_to_air can be calculated by summation of ind_fraction_CW_OT_to_air and ind_fraction_CW_R_to_air, both in “Source 3”. Then, using such calculated indicator and by reverting the formula for ind_water_to_water, it is possible to derive ind_fraction_TW_to_air and its equivalent indicator ind_fraction_PW_to_air from “Source 1”, “Source 2”. Given that all the needed indicators are either available or a derivation from the available sources has been calculated, the target indicator ind_water_to_air can be finally calculated. Conversely, if the enterprise data sources cannot provide the necessary data to compute a target indicator, the service can assist in identifying the missing indicators. This functionality is crucial for organizations to understand their data gaps better and enhance their data collection and management policies.
A suite of more advanced reasoning services can be used for facilitating collaboration among a network of enterprises, enabling them to more easily perform consistent sustainability assessments. This is particularly important for business processes spanning multiple enterprises, in which the overall assessment requires the parties to share relevant indicators in order to calculate the impact indicator at hand. A set of reasoning services are able to determine what indicators are computable starting from those already available, and what are the common dependencies for the target impact indicators to assess. On its top, a dependency calculation service allows to compare the available indicators for each collaborating enterprise, in order to determine the set of common (available or computable) indicators among them, namely the indicators that can be used for making assessments at network level.
This article presents an approach based on a formal, shareable representation for LCA data from the Ecoinvent database, which can facilitate the computation and comparison of related indicators and the formal verification of the compliance in evaluating LCIA. By leveraging Knowledge Graphs, our framework supports mapping internal data sources to LCA indicators, thereby aiming to improve the accuracy and reliability of the assessments, while reasoning services can be used for automatic identification of computable indicators and aggregation of basic indicators into compound indicators. This may provide valuable support for collaboration among enterprises, enabling joint sustainability assessments and efective decision-making. Future work will focus on evaluating the approach on real-world cases, expanding reasoning services, integrating additional LCI databases, and exploring applications in various industrial sectors.
Cristina Rossetti has received funding from the MUR – DM 118/2023 as part of the project PNRR-NGEU.
The author(s) have not employed any Generative AI tools.