paper.md

title	tags	authors	affiliations	date	bibliography
mescal: coupling energy system models with life-cycle assessment	Python energy system models life-cycle assessment	name orcid affiliation Matthieu Souttre 0000-0002-6288-058X 1 name orcid affiliation Guillaume Majeau-Bettez 0000-0002-0151-2468 1 name orcid affiliation François Maréchal 0000-0003-1752-5690 1, 2 name orcid affiliation Manuele Margni 0000-0002-2475-0768 1, 3	name index CIRAIG, École Polytechnique de Montreal, 3333 Queen Mary Road, Montréal, H3V1A2, Québec, Canada 1 name index Industrial Process and Energy Systems Engineering group, École Polytechnique Fédérale de Lausanne, Rue de l’Industrie 17, Sion, 1950, Switzerland 2 name index CIRAIG, Institute for Sustainable Energy, University of Applied Sciences Western Switzerland, Rue de l’Industrie 23, Sion, 1950, Switzerland 3	13 January 2025	paper.bib

Summary

Energy System Models (ESM) are widely used to design and assess energy transition scenarios. They help decision-makers to project their policies and understand their impact on strongly interconnected energy systems. However, ESM typically lack environmental indicators, which are essential to assess the sustainability of energy transition scenarios. mescal is a Python package that integrates Life-Cycle Assessment (LCA) indicators within ESM. It allows energy modellers to include a wider set of environmental indicators within their models in a transparent and reproducible way. As a result, energy modellers and decision-makers can identify transition scenarios trade-offs and hot spots, thus enabling a more holistic sustainability assessment.

Statement of need

ESM aims to support decision-makers to design and assess energy transition scenarios. In this work, we focus on bottom-up models, which describe energy technologies with a high level of technical details [@herbst2012]. Most ESM integrate carbon emissions in their modelling framework to propose and assess climate change mitigation scenarios. However, the environmental assessment of energy transition scenarios should not be limited to carbon emissions [@hellweg2023]. For instance, the United Nations highlighted three major environmental challenges that humanity is currently facing: climate change, biodiversity loss, and air pollution [@unfccc2022]. Sustainability assessment methods are needed to integrate a broader set of environmental impact categories, including as water use, land use, human toxicity, mineral resource use, etc. It is essential to enlarge the environmental assessment of energy transition scenarios while keeping comparable, robust and transparent methods [@hellweg2023].

LCA is a methodology that assesses the environmental impacts of products and services throughout their life-cycle. It aims to be a comprehensive environmental assessment as it considers complete value chains and a wide set of environmental impact categories.

ESM can be coupled to LCA with varying integration levels. Ex-post analyses are the most common approach, where energy transition scenarios generated by ESM are assessed with LCA as a post-processing step [@fernandezastudillo2019; @blanco2020; @junne2020]. Soft-linking approaches further use LCA results to adjust ESM projections, thus leading to iterative feedback loops between the ESM and the LCA model. For instance, @xu2020 computed environmental impacts of the energy mix obtained from an ESM, and then adjusted energy transition pathways when policy targets were violated. For LCA indicators to be used actively in ESM, and thus in transition scenarios design, hard-linked coupling is required. This denotes the endogenous integration of LCA indicators in ESM. Several studies have endogenously integrated LCA indicators in ESM [@rauner2017; @vandepaer2020; @algunaibet2019; @reinert2022], thus paving the way for multi-objective optimization. @vandepaer2020 compared the use of the $\epsilon$-constraint method and the normalization and weighting method to integrate LCA indicators in a bottom-up optimization model. Alternatively, @algunaibet2019 monetized environmental impacts in order to sum them with direct and indirect costs within an economic objective function.

However, these studies lack generalization and are hardly reproducible when considering different ESM.

Recently, @sacchi2024 presented pathways, a Python package based on the brightway framework [@mutel2017] that assesses energy transition scenarios with LCA. Given an energy transition scenario generated by an ESM or Integrated Assessment Model (IAM), prospective LCI databases (which are typically generated using premise [@sacchi2022], a Python package that generates prospective versions of ecoinvent using IAM projections) and a mapping between ESM/IAM scenario variables and LCI datasets, pathways computes the environmental impacts of transition scenarios and provides insights about their trade-offs and hot spots. pathways allows a transparent and reproducible assessment of energy transition scenarios. ESM/IAM results can be further included in a prospective LCI database using premise custom scenarios [@sacchi2022], thus making soft-linked coupling possible. However, this framework does not allow for the endogenous integration of LCA indicators within ESM or IAM, thus significantly limiting energy modellers ability to shape transition scenarios with LCA results.

mescal aims to address this limitation, by enhancing the reproducibility and transparency of hard-linked coupling between LCA and ESM, thus allowing a better comparability among energy modellers envisioning to enlarge their set of sustainability metrics.

Description

mescal requires four mandatory CSV files:

• Mapping.csv: the mapping between ESM technologies/resources and LCI datasets from a LCI database, e.g., ecoinvent [@wernet2016].

• ESM.csv: the input and output energy vectors of ESM technologies, e.g., a gas boiler with 90% efficiency has an input of 1.11 kWh of natural gas and produces 1 kWh of heat.

• Conversion factors.csv: the set of unit conversion factors between ESM technologies/resources and their mapped LCI datasets, e.g., from GWh to MJ.

• CPC.csv: the mapping between ESM energy vectors (e.g., electricity, heat, natural gas) and Central Product Classification (CPC) categories [@unitednations2015].

In addition, mescal can use several optional CSV files, including:

• Lifetime.csv: the lifetime of the ESM technologies in the ESM and in the LCI database.

• Efficiency.csv: the list of ESM technologies and their input fuel type for which efficiency adjustment will be performed.

• Ranking locations.csv: the ranking of preferred locations available in the LCI database with respect to the ESM geographical context.

mescal operates a set of systematic transformations (\autoref{fig:workflow}) on LCI datasets based on the brightway framework [@mutel2017] and the wurst Python package, which are tools to perform LCA modelling and calculation within a Python environment.

Mapping between ESM technologies/resources and LCI datasets

Each technology or resource of the ESM is mapped with one or several LCI datasets (Mapping.csv). Technologies are typically mapped with two LCI datasets: 1) an operation LCI dataset, which encompasses the use phase of the technology's life-cycle, and 2) an infrastructure LCI dataset, which encompasses the construction and dismantling phases of the technology's life-cycle. ESM resources (which can also be seen as energy imports) are technology inputs. They are typically mapped with one operation LCI dataset, which encompasses the resource extraction, processing, and transportation phases. Regarding LCI databases, mescal is suited for any version of the ecoinvent database [@wernet2016] and any prospective LCI database generated via premise [@sacchi2022].

Regionalization and spatialization of LCI datasets

To make LCI datasets more specific and relevant to the ESM geographical scope, the inventory can be regionalized. Inventory regionalization refers to the enhancement of the geographic representativeness of the inventory data, which is achieved by adjusting the type and quantity of intermediary and elementary flows [@patouillard2018]. mescal regionalizes the foreground inventory by modifying the geographical location of the flows of all infrastructure and operation LCI datasets according to the ranking of most suitable LCI database locations with respect to the ESM geographical context, Ranking locations.csv (\autoref{fig:flowchart_regionalization}). Foreground regionalization is always performed for intermediary flows, and also for elementary flows if the LCI database is spatialized, i.e., if a geographic location has been attributed to elementary flows [@patouillard2018]. The types and quantities of the flows remain unchanged. The LCI database can be spatialized and its background inventory can be regionalized using the regioinvent library [@maximeagez2025].

{ width=75% }

Double-counting removal

Double-counting refers to the multiple occurrence of energy flows within the energy system supply chain, thus leading to an overestimation of environmental impacts. For instance, if a coal power plants produces electricity that is later used in a heat pump, impacts of the electricity flow are counted twice: once in the coal power plant LCI dataset and once in the heat pump LCI dataset (input intermediary flow of electricity). @volkart2018 proposed a solution to the double-counting issue by setting to zero all flows in the foreground inventory that were also modelled in the ESM. This approach is implemented in mescal by identifying the flows to be nullified using their CPC categories [@unitednations2015] (\autoref{fig:flowchart_double_counting}). In the previous example, the input electricity flow of the heat pump operation LCI dataset would be set to zero.

In the case of market-type LCI datasets, the flows to be nullified are contained in the background inventory. mescal determines the set of LCI datasets on which the double-counting removal process should be applied via a recursive algorithm exploring the market-type datasets backgrounds (\autoref{fig:flowchart_background_search}).

ESM and LCI database harmonization

mescal adjusts LCI datasets and specific impact scores to account for differences between the ESM and LCI database:

• Technologies lifetime: Infrastructure LCA indicators are annual impacts, thus mescal adjusts the infrastructure specific impact scores to integrate the difference of lifetime between ESM technologies and their infrastructure LCI datasets. The infrastructure specific impact score ($lcia_{infra}$) is multiplied by the ratio between the ESM lifetime ($n_{ESM}$) and the LCI dataset lifetime ($n_{LCI}$) (Lifetime.csv) to ensure that the annual impact in the ESM is computed with the LCI dataset lifetime, thus resulting in the adjusted infrastructure specific impact score ($lcia_{infra}^{adj}$ in Eq. (1)).

$$ lcia_{infra}^{adj}(j,k) = lcia_{infra}(j,k) \cdot \frac{n_{ESM}(j)}{n_{LCI}(j)} \quad \forall (j,k) \in TECH \times ENV \quad \text{(1)} $$

• Technologies efficiency: Efficiencies of technologies in the ESM and LCI database should be harmonized, even if input fuel flows are set to zero in the operation LCI dataset to prevent double-counting, because a difference in efficiency between a technology and its corresponding operation LCI dataset would result in an inconsistency regarding the amount of direct emissions. mescal resolves this issue by adjusting the amount of direct emissions proportionally to the efficiency difference. Except land occupation, land transformation and energy elementary flows, the amounts ($q$) of all elementary flows in the operation LCI datasets foregrounds are adjusted using the ratio between the LCI dataset ($\eta_{LCI}$) and the ESM ($\eta_{ESM}$) efficiencies, thus resulting in adjusted direct emissions amounts ($q^{adj}$ in Eq. (2)). The efficiency of the operation LCI dataset ($\eta_{LCI}$) is computed using the quantity of fuel that was removed during the double-counting removal step, while the efficiency of the ESM technology ($\eta_{ESM}$) is computed from ESM.csv. This transformation is applied to a list of relevant ESM technologies (Efficiency.csv), e.g., technologies that involve a combustion process.

$$ q^{adj}(ef, j) = q(ef, j) \cdot \frac{\eta_{LCI}(j)}{\eta_{ESM}(j)} \quad \forall (ef, j) \in EF \setminus {\text{land, energy}} \times TECH \quad \text{(2)} $$

• Physical units: The product flows may be expressed in different units in the ESM and the LCI database. Specific impact scores are multiplied by a conversion factor, which converts the specific impact scores physical unit from [impact category unit / LCI output unit] to [impact category unit / ESM output unit]. Conversion factors encompass LCI datasets assumptions such as capacity factors or vehicle load factors (Conversion factors.csv).

• Generation of new LCI datasets: mescal generates new LCI datasets by operating modifications on existing ones, to enhance their alignment with the corresponding ESM technology. For example, biodiesel-fuelled mobility LCI datasets are created by replacing direct fossil carbon emissions by biogenic carbon emissions, while the fossil diesel input is set to zero during the double-counting removal step.

Life-Cycle Impact Assessment

mescal can compute LCA indicators using any set of impact assessment methods, e.g., IMPACT World+ [@bulle2019], ReCiPe [@huijbregts2017] or Environmental Footprint (EF) [@europeancommission.jointresearchcentre.2018]. Alternatively, a module computing only direct emissions has been developed to ease the comparison between territorial emissions and life-cycle emissions. This module sets all foreground intermediary flows to zero, thus only considering direct emissions for the impact assessment. The equivalence between territorial emissions and direct emissions is based on the assumption that all modeled direct emissions are located in the geography of interest.

Normalization of impact scores

Prior to integration into ESM, specific impact scores are normalized. In the context of optimization, normalization is beneficial in facilitating the solver's convergence, given that specific impact scores may exhibit significant discrepancies in magnitude across impact categories and technologies. By aligning all metrics within a comparable order of magnitude, numerical stability is improved in the solving process. Furthermore, considerable discrepancies in magnitude may be observed between infrastructure and operation specific impact scores within the same impact category, as these are not expressed with the same physical unit (e.g., kg CO$2$-eq/kW for infrastructure and kg CO$2$-eq/kWh for operation). Consequently, a scaling factor ($lcia{op,max}(k) / lcia{infra,max}(k)$ in Eq. (4)) is applied to infrastructure specific impact scores, to ensure that both the highest infrastructure and operation indicators are normalized to 1. The scaling factor invert is then applied to normalized infrastructure indicators (Eq. (6)), in order to keep the magnitude difference between operation and infrastructure specific impact scores in ESM. Normalization is performed using the maximum indicator ($lcia_{max}$ in Eq. (3)) of each impact category. In addition, all normalized indicators ($lcia_{type}^{norm}$ in Eq. (6)) that are below a threshold ($\epsilon$) are set to zero. This aims to determine the maximum order of magnitude between the highest and lowest indicators of an impact category, to eventually facilitate the solver convergence.

$$ \begin{split} lcia_{type,max}(k) & = \max(lcia_{type}(j,k) \ | \ j \in TECH \ \cup \ RES) \\ & \forall type \in {infra, op} \quad \forall k \in ENV \quad \text{(3)} \end{split} $$

$$ lcia_{infra}^{scaled}(j,k) = lcia_{infra}^{adj}(j,k) \cdot \dfrac{lcia_{op,max}(k)}{lcia_{infra,max}(k)} \forall (j,k) \in TECH \times ENV \quad \text{(4)} $$

$$ lcia_{max}(k) = \max(lcia_{type,max}(j,k) \ | \ type \in {infra, op}, \ j \in TECH) \quad \forall k \in ENV \quad \text{(5)} $$

$$ lcia_{type}^{norm}(j,k) = \begin{cases} 0 \text{ if } \dfrac{lcia_{type}^{(scaled)}(j,k)}{lcia_{max}(k)} \leq \epsilon \\ \dfrac{lcia_{type}^{(scaled)}(j,k)}{lcia_{max}(k)} \cdot \dfrac{lcia_{infra,max}(k)}{lcia_{op,max}(k)} \text{ elif } type = infra \\ \dfrac{lcia_{type}^{(scaled)}(j,k)}{lcia_{max}(k)} \text{ else} \end{cases} $$ $$ \forall (j,k) \in TECH \ \cup \ RES \times ENV \quad \forall type \in {infra, op} \quad \text{(6)} $$

Equations specification

The following set of modelling equations is included in ESM. The environmental objective ${LCIA_{tot}}$ is defined as the sum of the impacts of the infrastructure, operation, and resource parts (Eq. (7)). The infrastructure impact is derived from the normalized specific impact ($lcia^{norm}{infra}$), which is computed from the infrastructure LCI dataset. The normalized specific impact is divided by the technology's lifetime in the ESM ($n{ESM}$), and scaled with the technology's installed capacity (${F}$) (Eq. (8)). The operation and resource impacts are respectively derived from the operation and resource normalized specific impacts ($lcia^{norm}{op}$), which are respectively computed from the operation and resource LCI datasets, and scaled with the annual operation (${F_t} \times t{op}$) (Eq. (9)).

$$ \begin{split} {LCIA_{tot}}(k) & = \sum_{j \in TECH} \left( {LCIA_{infra}}(j, k) + {LCIA_{op}}(j, k) \right) + \sum_{r \in RES} {LCIA_{op}}(r, k) \\ & \forall k \in ENV \quad \text{(7)} \end{split} $$

$$ {LCIA_{infra}}(j, k) = lcia_{infra}^{norm}(j, k) \cdot {F}(j) \cdot \frac{1}{n_{ESM}(j)} \quad \forall (j, k) \in TECH \times ENV \quad \text{(8)} $$

$$ {LCIA_{op}}(j, k) = lcia_{op}^{norm}(j, k) \cdot \sum_{t \in T} {F_t}(j, t) \cdot t_{op}(t) \quad \forall (j, k) \in TECH \cup RES \times ENV \quad \text{(9)} $$

Integrating ESM results in the LCI database

In order to update the LCI database with the ESM results, mescal overwrites the relevant LCI datasets, i.e., LCI datasets that are in the sectoral and geographical scope of the ESM, such as markets for electricity, heat or transport. The activities composing the market and their respective shares are determined using the ESM annual operation results and the Mapping.csv file .
Updating the LCI database background inventory paves the way for using mescal with myopic ESM, i.e., ESM dividing the transition period into a sequence of consecutive optimization problems [@prina2020], through an iterative 3-step procedure: 1) run the ESM at time-step $t$, 2) update the LCI database with the ESM results at time-step $t$, and 3) update the LCA indicators with the updated LCI database for time-step $t+1$.

An example notebook is available to illustrate the use of mescal.

Impact

mescal offers a systematic methodology to hard-link ESM with LCA, thus enabling the integration of environmental constraints and objectives in ESM, and thus a flexible design and environmental assessment of energy transition scenarios. The use of mescal ensures a transparent, reproducible, and thus comparable integration of LCA indicators in ESM. The use of LCA makes sustainability assessments of energy modellers more holistic, thus highlighting the potential trade-offs, benefits, and adverse side effects of energy transition pathways among the environmental and economic performance indicators. mescal is aimed to be used by energy modellers who might not be LCA experts but want to enlarge the set of environmental indicators in their model in a transparent and reproducible way.

As an example, mescal methodology has been applied by @schnidrig2024 with the EnergyScope model [@moret2017] to analyse environmental-economic trade-offs in Swiss energy system transitions.

Conclusion

mescal makes the integration of LCA indicators in ESM more transparent, reproducible, and comparable. It aims to encourage energy modellers to consider a broader set of environmental indicators in their energy transition scenarios, thus enabling a more holistic sustainability assessment.

Acknowledgements

The authors gratefully acknowledge the financial support of the Fonds de recherche du Québec - Nature et Technologies, the Institut de l’énergie Trottier de Polytechnique Montréal and the CREATE-SEED program.

Nomenclature

Sets

Symbol	Description
$TECH$	Set of ESM technologies
$RES$	Set of ESM resources
$T$	Set of time periods
$ENV$	Set of environmental impact categories

Parameters

Symbol	Description	Unit
$t_{op}(t)$	Time period $t$ duration	hours
$n_{ESM}(j)$	Lifetime of a technology $j$ in the ESM	years
$n_{LCI}$	Lifetime of a technology $j$ in the LCI dataset	years
$lcia_{infra}(j,k)$	Infrastructure specific impact of technology $j$ for impact category $k$	impact category unit / capacity unit
$lcia_{op}(j/r,k)$	Operation specific impact of technology $j$ or resource $r$ for impact category $k$	impact category unit / operation unit
$\eta_{LCI}(j)$	Efficiency of technology $j$ in the operation LCI dataset	dimensionless
$\eta_{ESM}(j)$	Efficiency of technology $j$ in the ESM	dimensionless
$q(ef, j)$	Amount of elementary flow $ef$ in the operation LCI dataset of technology $j$	elementary flow unit
$\epsilon$	Threshold for the normalization of LCA indicators	dimensionless

Variables

Symbol	Description	Unit
$F(j)$	Installed capacity of technology $j$	capacity unit
$F_t(j,t)$	Operation of technology $j$ in time period $t$	operation unit
$LCIA_{tot}(k)$	Normalized total impact of impact category $k$	dimensionless
$LCIA_{infra}(j,k)$	Normalized infrastructure impact of technology $j$ for impact category $k$	dimensionless
$LCIA_{op}(j/r,k)$	Normalized operation impact of technology $j$ or resource $r$ for impact category $k$	dimensionless

References