On the Potential of District-Scale Life Cycle Assessments of Buildings

Abstract

Climate neutrality goals in the building sector require a large-scale estimation of environmental impacts for various stakeholders. Life Cycle Assessment (LCA) is a viable method for this purpose. However, its high granularity, and subsequent data requirements and effort, hinder its propagation, and potential employment of Machine Learning (ML) applications on a larger scale. The presented paper outlines the current state of research and practice on district-scale building LCA in terms of standards, software and certifications, and data availability. For this matter, the authors present the development and application of two district-scale LCA tools, Teco and DisteLCA, to determine the Global Warming Potential (GWP) of three different residential districts. Both tools employ data based on (including, but not limited to) CityGML, TABULA, and ÖKOBAUDAT. The results indicate that DisteLCA’s granular approach leads to an overestimation of environmental impacts, which can be derived from the statistical approach to operational energy use and related emissions. While both tools lead to substantial time savings, Teco requires less manual effort. The linkage of the aforementioned data sources has proven laborious and could be alleviated with a common data framework. Furthermore, large-scale data analysis could substantially increase the viability of the presented approach.

1. Introduction

The ambitious goals for climate neutrality defined in the Paris Agreement of 2015 [1] have led to national commitments to reduce CO${}_2$ emissions, e.g., the German Climate Neutrality Act of 2021 [2]. Such regulations often stipulate decarbonisation objectives for specific industries. In the built environment, emissions occur from the supply of heat, cooling, and electricity during the operational stage. Furthermore, the production of building components and assembly thereof, as well as the dismantling and disposal of such components, have environmental effects, including CO${}_2$ emissions. Thus, the full decarbonisation of the building sector requires the consideration of all building life cycle stages. The transition from the use of fossil fuels to renewable sources of energy is leading towards the decentralisation of power generation. Renewable energy systems such as geothermal, solar, or wind power plants can be integrated into the vicinity of districts to locally supply heat, cooling, and electricity [3]. Therefore, a district-level view of the building sector is sensible to determine the energy supply and demand of buildings and subsequent emissions, including the emissions from other life cycle stages. This is particularly relevant for urban planners in the planning process [4], policymakers for the allocation of building and refurbishment incentives [5], and real estate investors to comply with Environmental, Social, Governance (ESG) criteria [6] and the EU taxonomy [7]. Life Cycle Assessment (LCA) is a viable method for the determination of environmental impacts, including carbon footprints [8]. The high granularity and level of detail of this method presuppose a high data demand and computational effort, even more so with the simultaneous consideration of numerous buildings [9]. This raises the research question of to what extent a simplified method could approximate LCA results to a satisfactory extent while requiring less data input and computational time.

The aim of this research is to coherently outline the status quo and the potential of building LCA at a district scale. This includes normative boundary conditions, available datasets, programmes, and established building and district certification systems. Furthermore, current focal points in research are presented; in particular, data accessibility and synthetisation (see Section 1.1). In this context, this contribution presents the development and programme architecture of two district-scale LCA tools, Teco and DisteLCA (Section 2). Both tools are applied to a series of example residential districts with varying urban environments and building typologies in Germany (Section 2.3). Teco serves as a heuristic, i.e., approximative, LCA method for the calculation of buildings’ carbon footprints. DisteLCA; however, is the upscale extension of a highly granular single-building LCA software (version 0.9.7). For the purposes of this contribution, Teco is a heuristic tool for the carbon footprint determination of districts and will be applied to the aforementioned example districts. DisteLCA, however, serves as a granular comparison tool for the calculations of Teco. The strengths and current limitations of Teco’s and DisteLCA’s methodology are critically examined (Section 4), with a particular focus on each tool’s suitability for large-scale applications in data science. This contribution concludes with the derivation of further developments that are required to refine the methodology in terms of result accuracy and input simplification (Section 5).

1.1. Literature Review

This section provides an overview of the state of the art in the LCA of buildings and districts. Current standards and guidelines are described with respect to the general methodology and data requirements of LCA. The outline is limited to the key aspects with regard to the intentions of this contribution. Subsequently, the availability of data related to buildings and their environmental impacts is discussed. This is followed by a description of current building LCA software and the role of LCA in building and district certification systems. Furthermore, the focal points of ongoing research are outlined to conclude with the essential requirements to increase the scalability and accuracy of building and district LCA.

1.1.1. Standards and Guidelines

The general methodology of LCA has been standardised in ISO 14040 and ISO 14044 [10,11,12]. These international standards define the framework and key terminology to describe the respective processes (see Figure 1).

In particular, (i) the functional unit as part of the LCA goal defines the system whose performance is calculated, e.g., 1 m${}^2$ of a standardised dwelling. Furthermore, the reference flow is defined, which is the amount of a resource needed to produce the functional unit, e.g., 20 tons of concrete. (ii) The scope comprises the assessment parameters, such as GWP in terms of climate change, and the spatio-temporal boundaries of the system. (iii) The inventory analysis summarises the physical input and output flows, mainly resources, and products, in accordance with the reference flows. The inventory analysis results in the Life Cycle Inventory (LCI). (iv) The Life Cycle Impact Assessment (LCIA) translates the LCI into emissions and environmental impacts. This is achieved by employing impact categories suitable for the chosen assessment parameters, e.g., GlobalWarming Potential (GWP) for climate change, and allocating the LCI flows to these categories. (v) The interpretation of the results consists inter alia of sensitivity and uncertainty analyses to assess the quality of the LCA study. The bidirectional flows in Figure 1 emphasise the iterative nature of the overall LCA process. This is to consider practical limitations, mainly the overall data availability and quality [13]. ISO 14044 and ISO/TS 14048 have been established to complement ISO 14040 and to provide clarity over data requirements and guidelines for the conduction of LCA studies [12,15,16]. Furthermore, the European International Life Cycle Data System (ILCD) has been developed with a range of technical handbooks [17,18,19,20] and dataset documentation guides [21,22,23,24]. These documents complement the ISO framework with specific guidance to fulfil the ISO requirements, avoid ambiguities in the assessment process, and standardise the setup of LCA databases [13]. Sector-specific standards have been developed for the built environment. ISO 21930 and EN 15804 introduce the Environmental Product Declaration (EPD) as an indicator of the ecological quality of construction products. The standards stipulate a life cycle model consisting of stages A to D with respective stages and allocate them to different system boundaries as fulfilment standards for EPDs [25,26]. Figure 2 gives an overview of the aforementioned system. More standards such as DIN 15978 have been established to refine the system boundaries and to define scenarios for the temporal development of LCI flows [27,28]. Current German normative work on LCA with regard to buildings and districts recognises the potential of Machine Learning (ML)-based tools for the efficient determination of environmental impacts. It also acknowledges the necessity to produce machine-readable LCA data according to common data standards and semantics [29,30,31].

1.1.2. LCA Data Availability

The LCI of buildings and districts necessitates a broad range of information on materials and elements used in the construction of a building to determine the input and output flows (see Section 1.1.1). This information is usually saved in a geometric or Building Information Modelling (BIM)-based model. However, there is a general lack of publicly available, highly detailed building or district models due to data privacy concerns, and arduous data acquisition. [32]. Yet, Geographical Information System (GIS)-based buildings’ geometrical data are widely available. This includes City Geographical Markup Language (CityGML) [33], Geo Javascript Object Notation (GeoJSON) [34], ESRI File GeoDataBase (ESRI File GDB) [35], Geo Javascript Object Notation (GeoJSON) [36], and Shapefile [37]. CityGML is the most frequent encoding standard of publicly available GIS-based building data used in the field of Urban Building Energy Modelling (UBEM) [38], and is the focus of the presented research. The standard introduces a Level Of Detail (LoD) concept of buildings, whereby the amount of information in a CityGML model increases with the LoD of the buildings. Note that this contribution focuses on CityGML version 2.0, whereby version 3.0 has been published already [39]. Figure 3 gives an overview of the LoD concept [40]. Note that CityGML-based building data are generally available in LoD1 and LoD2 in Germany and Europe [33,38]. To enhance the usability of CityGML models for urban-scale building and energy network simulations, the Energy Energy Application Domain Extension (ADE) and Utility Networks ADE have been developed [41,42]. In particular, the current version of Energy ADE 2.0 extends the CityGML standard with classes and attributes related to building physics and patterns of occupancy [38].

Previous research has attempted to further extend the Energy ADE with a generic description of utilities, a reduced-order representation of energy networks, and differentiated information about LCA-related parameters. This includes renewable and non-renewable energy consumption for heating and electricity, as well as environmental impact categories [43,44]. The interest in the GIS-based district-scale LCA of buildings has been increasing [32,45]. However, to the knowledge of the authors, there is no publicly accessible tool or database that allows the allocation of LCI- and LCIA-related data to CityGML-based building geometries. Yet, there is a bandwidth of commercial [46,47] and publicly available [48,49,50] LCA databases for construction materials [51]. These databases provide datasets of building materials and compound products, with information about the reference flow of the item, the modelling and validation process of either generic or specific datasets, and LCI input and output flows of energy consumption and environmental impacts. In Germany, the publicly available ÖKOBAUDAT database [48] provides such information in the form of EPD in accordance with DIN EN 15804 [26] and ILCD modelling guidance.

1.1.3. Software and Certification Systems

A broad range of generic and building-specific LCA software exists. Such programmes allow the integration of independent or proprietary LCA databases (e.g., openLCA [52] and Gabi [53], respectively). The tools may be either stand-alone software (e.g., LEGEP [54]), plugin-based (e.g., OneClickLCA [55]), or web-based (e.g., eLCA [56]). Some of these applications allow for the import of BIM or Industry Foundation Classes (IFC) building models [57]. The tool DisteLCA (see Section 1) is an extension of eLCA.

Sustainable building and district certification systems emphasise environmental quality and are a significant contributor to the dissemination of LCA as a method to determine the overall building quality [58]. The largest market participants in Europe are the German Sustainable Building Council (Deutsche Gesellschaft für Nachhaltiges Bauen (DGNB)), Leadership in Energy and Environmental Design (LEED), and Building Research Establishment Environment Assessment Method (BREEAM). Their market shares vary significantly in Europe and Germany, with DGNB being the most prevalent system in Germany [59,60]. The DGNB system offers building and district certifications for different usage types. It stipulates the conformity of the involved LCA with the EPD regulations of DIN EN 15804 [26], i.e., the consideration of life cycle stages A to D, and the general Life Cycle methodology according to DIN EN ISO 14040/14044 [14,15]. Furthermore, the operational energy demand is to be determined according to the procedures of DIN V 18599 for the calculation of net, final, and primary energy demands [61]. DGNB suggests the usage of the LCA database ÖKOBAUDAT. Also, the conformity with the German Buildings Energy Act [62] has to be verified, and the architectural and utility components of the buildings have to be categorised according to the cost groups of DIN 276 [63,64,65]. The LEED certification system follows a similar rationale except for the requirement to conform with ISO 21930 [25] instead of DIN EN 15804 [66,67,68,69]. BREEAM is an analogous system, including requirements for the LCA tool compliance [70]. The German Assessment System for Sustainable Building (Bewertungssystem Nachhaltiges Bauen (BNB)) adds eLCA as an exemplary LCA tool [71]. There is a strong similarity between the sustainability parameters considered in each certification system, albeit the weighting of the aspects differs [72]. To the knowledge of the authors, all presented certification systems require the LCA calculation to be compliant with ISO 14040, ISO 14044, ISO 21930 or DIN EN 15804, and EN 15978, i.e., the systematic approach described in Section 1.1.1, with the evaluation of all impact categories across EPD system boundaries. Furthermore, all systems require a high degree of granularity in constructive building elements and allow for conservative estimations of the environmental impact of building services in life cycle stages A, C, and D.

1.1.4. Current Focal Points in Research

The research interest in LCA has been increasing steadily since the beginning of the 21st century [8]. In particular, building LCA research is increasing and focuses on the determination of energy demands and carbon footprints [73,74,75]. The challenges identified in this context consider the general data intensity and shortage, the uncertainty of static and dynamic LCA applications, the choice of proper system boundaries, inconsistent functional units, and the lack of a common modelling framework for building stock and LCA tools [32,75,76]. In particular, on the district scale, data availability in early design stages and the definition of a functional unit are ongoing challenges [9]. Moreover, building stock modelling suffers from a varying spatio-temporal scope, an unclear quantification of re-usage across system boundaries with a subsequent hazard of double counting values of stage D, and a lack of studies with highly detailed building models [77,78]. A meta-analysis on building LCA studies conducted by Bischof et al. [76] concludes that most building stock modelling approaches evade the widespread bottom-up/top-down categorisation, rather showing characteristics of both categories. However, archetype assignment is the most common approach for data enrichment in UBEM [79,80]. In the context of this research, archetypes are statistical renditions of building properties that are assigned by a limited range of input variables, in particular building age classes and usage types. These building representations are frequently used in UBEM to accelerate the processes of data collection and computation [81].

Yet, archetypes are generally constructed on the building rather than the urban level, which hinders the consideration of district-scale effects such as energy supply or transportation within the spatial boundaries. Furthermore, the current archetype approach cannot fully diminish the lack of data in LCA [82]. Even more so, current data deficiencies complicate the definition of accurate district LCA archetypes [83]. To overcome such deficiencies and to synchronise various data sources, Fnais et al. [84] suggest the development of a common data model for dynamic BIM, GIS, and LCA data with the inclusion of data semantics and information about the temporal development of material flows and environmental indicators. Feng et al. [74] contend that building LCA is currently experiencing a bottleneck period in which missing data about building structures and system boundaries results in broad uncertainties and, thus, limits the potential of the method. Their study suggests a framework to address such uncertainties by allocating functional units, i.e., building elements, to an uncertainty taxonomy, and to apply statistical methods to identify the most significant influential factors on the LCI in the future. Mastrucci et al. [85] developed a spatio-temporal LCA framework for district-scale refurbishments, and applied it to a district use case in Luxemburg. The case study introduces the district-scale functional units of “Carbon footprint reduction potential ($\frac{{k{t}_{CO}}_{2eq}}{ha·y}$)” and “Carbon footprint intensity reduction potential ($\frac{{k{g}_{CO}}_{2eq}}{m^2·y}$)”, and emphasises the lack of urban-scale data on building materials and service life, refurbishment rates, and the operational stage to increase the accuracy of their results. This is stressed by studies where LCA uncertainty analyses yielded the chosen building material type, the average service life, the share of renewable primary energy sources, and the energy distribution system as the most significant variables [86,87]. GIS data is being increasingly used for the sustainability assessment of districts [3,88,89]. In particular, Ang et al. [90] developed a platform (“ubem.io”) for the GIS-based determination of energy demands and the decarbonisation potential of a scalable number of buildings. Research on building and district LCA focuses increasingly on the potential of Machine Learning (ML) applications to narrow the uncertainty of variables, and to increase the accuracy of LCA results [82,84]. On the building level, ML models are mainly established to determine energy consumption and GWP, with Artificial Neural Networks (ANN) being the most used method [91]. D’Amico et al. applied different ML algorithms on a parametrised non-residential building model to determine the energy demand and GWP, showing a high degree of approximation of the ML results by comparison to values obtained from granular simulations [92]. Pomponi et al. applied different ML algorithms to determine the GWP of concrete and steel structures, leading to the development of a mass and carbon footprint estimation tool with statistical uncertainty statements in the form of probability density functions [93]. In a literature review on ML in LCA, Ghoroghi et al. [94] infer that on the district and building level, ML is mostly employed at the LCI stage to predict missing data and to optimise the material selection in the design process. It is concluded that the standardisation of data reporting and the application of data dimensionality reduction methods will result in more efficient employment of ML in LCA. The authors underline the general conflict area between the high data demand of both LCA and ML algorithms, and the necessity of gap-filling methods to increase the data availability. Lastly, the usage of dynamic data may aid in the propagation of ML-based LCA.

2. Materials and Methods

The central issues of current LCA techniques include a general lack of data and no available common framework for building data and LCA tools. Yet, CityGML is the most prevalent, publicly available building and district data format in Germany and Europe [38], and open source LCA databases exist, e.g., ÖKOBAUDAT (see Section 1.1). Moreover, the considerable manual and computational effort of granular building LCA challenges the goal of highly detailed analyses [9]. This leads to the authors’ motivation for the development of more coarse tools, focusing on the LCA results’ comparability between different districts. Thus, the authors suggest the development of an enrichment framework that enables the combination of data from CityGML files and from ÖKOBAUDAT to heuristically determine the LCIA of scalable districts. In this context, the authors present the tool “Teco” as the extension of a preexisting heuristic UBEM framework. Teco takes CityGML LoD1 or LoD2 datasets as input to calculate a range of environmental indicators. For the assessment of Teco’s heuristic calculation methods, the authors also present the tool “DisteLCA” as the upscaled version of a granular, single-building LCA tool. The aforementioned tools Teco and DisteLCA are applied on three different example residential districts (see Section 2.3 and Section 3). In light of a general lack of standard accuracy in building LCA studies [75], the authors explicitly address the conformity with (i) ISO 14040, (ii) ISO 21930, and (iii) EN 15804. This considers (i) the application of the life cycle stages and definition of respective parameters (see Figure 1, Section 1.1.1), and (ii, iii) the compliance of the available input datasets and the tools’ output with Cradle-to-Gate life cycle stages and EPDs of construction products and services (see Figure 2, Section 1.1.1).

2.1. Tool 1: Teco

This section describes Teco [95], a heuristic approach to determine the life cycle inventory and assess the environmental impact of large building stocks implemented in Python and Modelica. It is based on the Tool for Energy Analysis and Simulation for Efficient Retrofit+ (TEASER+) [40,96], an enrichment framework for UBEM, and therefore extends its classes for LCA calculations. Teco and TEASER+ are available as open source under the MIT license. (https://github.com/RWTH-E3D/Teco/tree/EnergiesSpecialIssue23 and https://github.com/RWTH-E3D/TEASERPLUS, accessed on 18 July 2023).

TEASER+ is a tool for parameterising and generating thermal building models. It is a further development of the tool TEASER [96]. The tool combines geometric data from CityGML datasets with statistical data on building construction and usage. It considers various temperature and solar radiation parameters. The tool employs invariables for external convective heat transfer; thus, not considering wind speed or direction. TEASER does not consider shading and radiation on opaque surfaces for performance reasons. TEASER+ provides a database of building archetypes with predefined typical building constructions (walls, ceilings, floors, ground floors, rooftops, and doors) from the TABULA project [97], use conditions for thermal zones, and estimation factors for the up- or downscaling of building geometries [98]. These archetypes are categorised by year of construction and building usage. The structure of TEASER+ allows the addition of various archetypes for specific applications. After selecting suitable archetypes, the outer building elements are scaled to the corresponding building geometry size within the CityGML model. However, inner building elements, as well as windows and doors, are scaled based on the Net Leased Area (NLA) multiplied by predefined conversion factors. The buildings thus assembled are used by TEASER+ to parameterise Reduced Order Models (ROM) in the Modelica libraries AixLib [99], Buildings [100], BuildingSystems [101], or IDEAS [102]. Due to its statistical approach, TEASER+ is conceived to be accurate on a district scale rather than for single-building simulations. In a study of three different use cases, Remmen et al. [96] have demonstrated that TEASER’s heat load simulation output deviates by no more than 5.6 % from measured values on a district scale.

Teco extends TEASER+ for use in LCA of districts. Figure 4 gives an overview of Teco’s overall software architecture (version 0.5.8).

For this purpose, the building models and the enriched ROM from TEASER+ are utilised for LCI. The materials used in the district and their quantities are extracted from the TEASER+ building structure. Teco links service life data taken from the German Federal Institute for Research on Building, Urban Affairs, and Spatial Development (BBSR) (German name Bundesinstitut für Bau-, Stadt-, und Raumforschung) to model material and building element replacements throughout a given life cycle period. However, the tool incorporates a hierarchy of service lives depending on the respective material’s position in the building element. For instance, an external wall might have an external insulation layer with a service life of 20 years and an underlying moisture barrier with a service life of 15 years. In the case of the moisture barrier’s replacement, Teco assumes a replacement for the insulation as well. This is to consider practical refurbishment scenarios, where single replacements of inner material layers are unrealistic. Energy consumption during the use stage is estimated with the help of the heating load curve from the ROM, and power consumption is roughly estimated by the NLA of the buildings while domestic hot water is disregarded. For impact assessment, these LCI data are linked with environmental indicator datasets. The datasets included in Teco are formally based on the German standard DIN EN 15804+A1 [26], thus providing detailed indicators for the use of resources and environmental impact across all LCA stages (see Section 1.1.1). These EPDs originate from the ÖKOBAUDAT database version “OBD-2021-II” [48], which focuses on German building elements and materials. In summary, Teco semi-automatically calculates the characteristics of the environmental indicators for the district’s building materials, heating loads, and electricity consumption. Additional environmental impacts can be considered in Teco by manually adding other EPD datasets to the buildings in any quantity. In this way, for example, radiators or hot water tanks can be taken into account by adding adequate EPDs. Both TEASER+ and Teco are open-source and have been developed at the authors’ affiliated Institute of Energy Efficiency and Sustainable Building e3D at RWTH Aachen University. A detailed description of Teco’s development process and scientific validation thereof can be found in [95].

2.2. Tool 2: DisteLCA

DisteLCA is a Python-based tool that facilitates the creation of multiple projects in the eLCA component editor (German name: eLCA Bauteileditor) version 0.9.7 [103]. The eLCA component editor is a web-based software to generate LCAs of buildings administered by the BBSR. DisteLCA enables the assessment of exterior walls, windows, roofs, ceilings, inner walls, foundations, and heat supply systems, as well as the operational energy use for heating, hot water generation, and lighting. This includes all environmental indicators available in ÖKOBAUDAT, including GWP. DisteLCA targets time savings in using eLCA for districts without compromising eLCA’s fidelity of the results for individual buildings. Therefore, eLCA can be used to produce granular district LCA. The eLCA component editor employs the ÖKOBAUDAT dataset [48].

DisteLCA evolved from another eLCA automation developed by the authors, comparing the impact of various renovation options for different building archetypes on a district level. A programme modification was necessary, as district LCA focuses on analysing existing buildings rather than comparing remediation alternatives. DisteLCA has been developed by some of the authors involved in this contribution and is available open-source. The underlying architecture of eLCA (or “bauteileditor”) is the result of a research project conducted by BBSR [104,105]. Figure 5 illustrates a UML activity diagram of the developed automation.

The first phase is data entry. To initiate DisteLCA, a user has to submit eLCA login credentials. While a standard account can contain up to 15 projects, the authors were supported with an unlimited account by the BBSR during the DisteLCA development. Before launching the programme, users of DisteLCA must add the specific components of the district under investigation to the account templates on the website. The component materials have to be compliant with the ÖKOBAUDAT version 2021 II A1. DisteLCA accesses the provided account to read out all private templates on eLCA outer walls, inner walls, ceilings, roofs, foundations, windows, and heating supply systems. In the next step, the user can enter information on the different building archetypes of the district to be analysed via the Graphical User Interface (GUI).

For each archetype, the user indicates:

• The number of buildings of that archetype in the district;
• The net and gross floor areas;
• The final energy demands for heating, hot water generation, and lighting;
• The areas of all building components included in the analysis;
• The corresponding eLCA component templates to be selected from drop-down menus;
• The energy source;
• The building age class.

For the presented research, final energy demands are derived from the TABULA typology. These values are national averages for the respective building archetype and are thus insensitive to a building’s actual location. DisteLCA processes the user input to a CSV file and a JavaScript Object Notation (JSON) file for each archetype. Thereafter, DisteLCA sends the data to the eLCA web server, and the projects on the district archetypes are created automatically through CSV import in the specified eLCA account. After the successful creation of all projects in eLCA, the data output phase begins. The LCI and LCIA evaluations of the generated projects are read from eLCA and compiled. A period of 50 years is observed. By multiplying the single building results for the LCI and LCIA by the number of buildings in the district, data is upscaled to the district level. All result tables are stored in CSV files in the district reports directory. DisteLCA’s LCI includes an overview of the building operation, a list of all materials and corresponding masses, and a summary of the user input data. The results for the life cycle impact assessment in DisteLCA give an overview of the summarised GWPs for materials and building operation, the results for the different life cycle stages, and all building materials and components. Finally, in order to restart DisteLCA, the user must delete all data in the DisteLCA district reports and temp data directories, including the eLCA projects. It is crucial to consider the limits of eLCA when interpreting the DisteLCA results and comparing DisteLCA to Teco. The eLCA component editor uses the information provided in the ÖKOBAUDAT. Contrary to Teco, eLCA, and DisteLCA do not fill in generic data for unspecified life cycle stages in the ÖKOBAUDAT. Furthermore, on the one hand, the usage of the eLCA component editor implies a high demand for data. On the other hand, this produces highly detailed results. Moreover, the necessary components must be available in the template collection; thus, if public templates are insufficient, more templates must be created manually in the component editor. However, compared to the non-automated usage of the eLCA component editor, time savings can be realised with DisteLCA. It takes one to two minutes per archetype to complete the DisteLCA GUI form if all required data are available. An example case with 14 archetypes results in up to 30 min of manual user input. Then, DisteLCA claims six to seven minutes to create 14 projects in eLCA and assemble the CSV output of the LCI and LCIA results on a single building and district scale. Creating a project by manual input takes around five minutes in eLCA, which makes up to 70 min for 14 projects. In this case, DisteLCA realises a time saving of 50% to 70%, increasing with a rising number of archetypes. In addition, upscaling the results to the district level and compiling them into CSV files provides added value in comparison to the manual use of eLCA and enables the targeted granular district LCA.

2.3. Application

The aforementioned tools Teco and DisteLCA are applied to three different example residential districts. These cases were selected across different regions of western and northern Germany. They represent varying climatic conditions, building densities (Single Family House (SFH) vs. Multi Family House (MFH)), building age classes, refurbishment statuses, and use of local building material characteristics. Note that project-specific agreements on data handling hinder the full disclosure of the districts. This includes their exact location and detailed architectural plans. The following description of each district contains postal code and nearest city references for climatic conditions, which are taken from Test Reference Year (TRY) [106] files in the weatherdata folder in the repository (https://github.com/RWTH-E3D/Teco/tree/EnergiesSpecialIssue23, accessed on 18 July 2023). The building element layers used by Teco can be found in the JavaScript Object Notation (JSON) files of the inputdata folder in the same repository. Note that these are fundamentally based on the TABULA building typology, which offers a web-based visualisation of building and system data per building age class (https://webtool.building-typology.eu/#bm, accessed on 18 July 2023). The layer setup in DisteLCA is listed in the appendix and available as Extensible Markup Language (XML) files for import in the “bauteileditor” (https://www.bauteileditor.de/, accessed on 18 July 2023) in the eLCA_example_archetypes folder in the repository (https://github.com/RWTH-E3D/DisteLCA/tree/EnergiesSpecialIssue23, accessed on 18 July 2023). In light of a general lack of standard accuracy in building LCA studies [75], the authors explicitly address the conformity with ISO 14040, ISO 21930, and EN 15804.

Table 1. Application of the ISO 14040 framework on the present scope.

Goal	Functional Units	$\frac{m_{NLA}^2}{distric{t}_i}$, $\frac{m_{NLA}^2}{buildingageclas{s}_j}$, $\frac{m^2}{buildingelemen{t}_k}$
	Reference Flows	Implicit by respective amount of building element/operational energy use k per building age class j per district i
Scope	Assessment Parameter	GWP (Climate change)
	Spatio-temporal boundaries	50 years in respective district and material transport in and out thereof
LCI	Input	Implicit via archetype definition of building elements and layers
	Output	Building elements by amount of houses in each building age class
LCIA	Impact Category	$GW{P}_{50a}$, expressed as $\frac{t_{C{O}_{2eq}}}{distric{t}_i}$, $\frac{t_{C{O}_{2eq}}}{m_{NLA,distric{t}_i}^2}$, $\frac{t_{C{O}_{2eq}}}{m_{buildingelemen{t}_k}^2}$, $\frac{t_{C{O}_{2eq}}}{buildingageclas{s}_j}$, $\frac{t_{C{O}_{2eq}}}{m_{NLA,buildingageclas{s}_j}^2}$
	LCI flow allocation	Trivial given single impact category

gives an overview of the applied method in accordance with ISO 14040 (Section 1.1.1).

The considered building elements in the LCI are roofs, external walls, foundations, internal floors/ceilings (referred to as ’floors’), windows, and utilities. All elements are expressed as m${}^2$ with the exception of utilities, which are considered in the operational stage.

Table 2. Data sources for the exemplary use of Teco and DisteLCA.

Element	Teco	DisteLCA
Roofs	CityGML file of district for building geometry, use type, and year of construction. TABULA archetype for buildings’ elements and materials. ÖKOBAUDAT datasets for environmental indicators of building materials and operational energy use.	CityGML file of district for buildings’ postal code, NLA, number of storeys, use type, and year of construction. Regionalised catalogue [107] of different construction types and material layers by postal code. TABULA for construction types and area-based heat energy demand per building archetype, where the regionalised catalogue is not applicable. ÖKOBAUDAT datasets for environmental impacts of building materials and operational energy use.
External walls
Foundations
Internal floors
Windows
Utilities	Gas condensing boiler of respective building age class in TABULA.

gives a succinct overview of the data sources and assumptions for the input in both Teco and DisteLCA. Where applicable, any information on buildings’ refurbishment status is only considered in the setup of the more granular tool DisteLCA since such project-related information is unavailable in CityGML files. This is achieved by using the usual or advanced refurbishment status of the TABULA archetype in case the project-related information indicates that such a refurbishment has been performed on the respective building.

Given the scope of climate change assessment for this LCA, all CO${}_2$ and equivalent emissions are determined. In the LCIA, the impact category $GW{P}_{50a}$ is expressed for life cycle stages A to D (Cradle-to-Gate, cp. Figure 2). Enrichment or interpolation of CityGML datasets to make the respective files usable for the aforementioned scope of LCA are explained where applicable. Furthermore, DisteLCA employs archetypes based on regional varieties of construction types [107], i.e., the prevalence and environmental implications of brick, wood, and concrete construction in the respective district are taken into consideration. These are complemented with assumptions on suitable material layers for flooring as well as wall and ceiling surfaces. The archetypes in DisteLCA exclusively use the Net Leased Area (NLA) of each basic archetype in the TABULA building typology [108]. To adapt the building element sizes of each building’s actual NLA, the estimation formulae of Loga et al. [98] are used for the evaluation of this research. Given the linear correlation between the mass or area of a building element and its GWP by its reference flow, the formulae are directly applied to the LCI and LCIA results. The environmental impacts of operational energy use are scaled analogously since they are founded on guideline values of the TABULA typology that are expressed per m${}^2$.

Example 1: Newly constructed residential district in Western Germany.

The first example is a newly constructed residential district in a rural area near the city of Cologne (postal code area 501XX), consisting of 104 SFH that are currently under construction and part of an ongoing research project. Appendix A.1 illustrates the materials per building element used in DisteLCA for this district. Figure 6 gives an overview of the buildings. The CityGML model is created using project-related information and CityBIT [109] since publicly available models do not yet exist.

Example 2: Existing residential district in Western Germany.

The second example district is a preexisting development of 134 MFH in an urban area near the city of Essen (postal code area 451XX) that were constructed between 1950 and 1961. The buildings were partially refurbished between 1990 and 2005. Appendix A.2 outlines the building materials as input in DisteLCA for this district. Figure 7 gives an overview of the buildings.

Example 3: Existing residential district in Northern Germany.

The third example is an urban district of 539 MFH constructed between 1918 and 2017 in the proximity of Hamburg (postal code area 200XX). Figure 8 shows an overview of the buildings. There is no information available about any refurbishments. Appendix A.3 lists the building materials employed in DisteLCA.

3. Results

This section describes Teco’s and DisteLCA’s output with regard to the functional units and impact categories (see Figure 1). The full raw output can be found in the repositories. (https://github.com/RWTH-E3D/Teco/tree/EnergiesSpecialIssue23 and https://github.com/RWTH-E3D/DisteLCA/tree/EnergiesSpecialIssue23, accessed on 18 July 2023). Values for stage D are depicted as absolutes in all figures and tables unless stated otherwise. Note that DisteLCA explicitly models life cycle module B4, whereas, in Teco, material replacements are considered in each relevant stage. Figure 9 gives an overview of the functional unit results. In alignment with the amount and types of buildings outlined in Section 2.3, district 3 shows the highest amount of NLA and building element area. It is also the only district to have buildings from all building age classes. In contrast, district 1 only has buildings from the newest typology class, and district 2 from classes 1949 to 1957, and 1958 to 1968. As could be expected, the large amount of MFH in urban district 3 leads to a high proportion of floor area.

The impact category results ($GW{P}_{50a}$) per life cycle stage are illustrated in

Table 3. $GW{P}_{50a}$ ($t_{C{O}_{2eq}}$) in each district per stage and $m_{NLA}^2$.

		Teco				DisteLCA
	District	A	B	C	D	A	B	C	D
$\frac{k{g}_{C{O}_{2eq}}}{distric{t}_i}$	1	5369	547	647	497	5871	10,748	1539	742
	2	4660	1317	2953	1,482	9218	37,434	2381	938
	3	176,659	21,201	30,992	19,248	116,451	1.7$·{10}^6$	15,052	5443
$\frac{k{g}_{C{O}_{2eq}}}{m_{NLA,distric{t}_i}^2}$	1	0.284	0.029	0.034	0.026	0.310	0.568	0.081	0.039
	2	0.100	0.028	0.062	0.031	0.194	0.787	0.050	0.020
	3	0.300	0.035	0.052	0.032	0.195	2.865	0.025	0.009

. Teco tends to output lower values for stage B. With the exception of district 1, Teco shows higher figures for stages C and D, with no conclusive trend for stage A. The subsequent emissions per $m_{NLA}^2$ per stage and district show different distributions in both tools.

Figure 10 provides an overview of $GW{P}_{50a}$ per building element. Note that stage D has been omitted for this representation since DisteLCA is incapable of issuing information on this model on a building element level. Furthermore, in light of DisteLCA’s results for stage B (see Table 3), the operational energy use is not part of the figure for visual clarity. Teco is prone to give higher outputs for all elements and districts, with the exception of roofs, and external walls in district 1.

Figure 11 shows the $GW{P}_{50a}$ per age class in each district. There is a clear tendency for Teco to give much lower results than DisteLCA for stages A to C (Figure 11a). However, the former tool generally determines a higher net benefit of stage D (Figure 11b). The GWP intensity per $m^2$ for each building age class and district is illustrated in Figure 12. For stages A to C combined (see Figure 12a,b), Teco outputs far lower results than DisteLCA. Moreover, both tools yield divergent peak GWP intensities for districts 2 and 3. A similar pattern can be observed for stage D (see Figure 12c,d), albeit Teco’s output values are mostly higher. The results show a general pattern. Teco tends to give higher estimates for stages A, C, and D separately, while DisteLCA tends to output lower values. However, DisteLCA produces far higher results for stage B than Teco for all examples. It should be noted that the ISO 14040 framework has been fully applied to the presented scope (see Table 1, Section 2.3). Yet, the EPD datasets provided by ÖKOBAUDAT sometimes lack specific stages of a Cradle-to-Gate scope such as modules C1 (Deconstruction/Demolition) and C2 (Transport). Analogously, neither tool produces output with regard to these modules. Teco’s distribution of replacements effectively omits phase B4 in the output.

4. Discussion

This section discusses (i) the modelling and simulation accuracy and usability of the presented application, (ii) data standardisation and availability, and (iii) the relationship between certification systems and LCA norms, and heuristic, large-scale building LCA tools.

4.1. Modelling and Simulation Accuracy and Usability

The application of both tools on the exemplary districts (see Section 2.3) has revealed output differences that can be explained by various, oftentimes counteracting effects. Teco tends to give lower estimates for life cycle stages A to C combined, and higher estimates for stage D than DisteLCA (Section 3). However, an in-depth view reveals that DisteLCA produces far higher results for stage B, whereas Teco tends to output higher values for modules A and C. The gap narrows significantly when only building elements (roofs, external walls, foundations, floors, and windows) are considered (see Figure 10). In this case, Teco’s results are slightly higher. This can be justified by Teco’s inclusion of replacements in stages A and C, as opposed to stage B4 in DisteLCA. Moreover, the usage of regionalised building archetype information in DisteLCA leads to different environmental impacts for building elements. However, this does not fully explain the extraordinary difference between the tools’ output for stage B. Thus, the tools’ most striking output difference is stage B6, i.e., the GWP from operational energy use. While Teco employs a dynamic heat load simulation approach (Section 2.1), DisteLCA upscales statistical values of energy use and GWP by $m^2$. With previous work indicating that Teco actually overestimates the GWP of operational energy use for certain districts [95], the authors conclude that DisteLCA’s statistical approach leads to a significant overestimation of emissions during buildings’ operational stage. This may be attributed to the fundamental difference between the dynamic simulation of the predicted energy demand in Teco, and the arithmetic means of actual energy use employed by DisteLCA. Given stage B’s significant influence on the overall results of DisteLCA, the tool’s GWP intensity output is distorted (see Figure 11). Thus, any order of GWP intensity by building age class should be rather established using Teco. However, it should be noted that the current version of the tool does not consider variable convection caused by wind conditions. Correspondingly, any manual LCA approach involving the same mathematical approach as DisteLCA is seen as disadvantageous for the determination of GWP intensity sequences.

The tool DisteLCA models building elements using regionalised information, with no conclusive influence on the output in the scope of the presented research. Furthermore, DisteLCA considers material layers for building furniture, e.g., floor coverings, resulting in higher GWP per building element. Teco distributes any element or material replacements among life cycle stages A to C, reducing the output of stage B by omitting module B4.

The usability of either tool for large-scale data production depends on their accuracy as well as computational and manual effort. With a large number of influential factors and their forecast nature, it is difficult to determine the precision of district-scale, or any, LCA a priori. It should be rather focused on creating comparability between different quarters, and the reproducibility of the results. For these reasons, a district-scale LCA tool is favourable in case it is able (i) to employ open-source input data, and (ii) to generate LCI and LCIA that can be traced back on this input data. Teco uses publicly available ÖKOBAUDAT environmental impact data, CityGML building models, and building typologies. While DisteLCA considers more material layers, the programme’s lack of native building archetype support implies substantial manual effort to consider a broad range of different building geometries and materials. As a consequence, Teco’s approach is seen as more favourable in generating LCA data on a large scale. However, it should be noted that the tool is not yet able to consider interior walls to an appropriate degree (see [95]). Moreover, any heuristic needs to be carefully calibrated in order to attain a satisfying balance between result accuracy and input as well as computational requirements. In this respect, Teco would benefit from more exemplary district use cases, and the development of similar workflows employing similar input data, i.e., CityGML.

4.2. Data Standardisation and Availability

The presented analysis of three different districts relies on publicly available data for building models and geometry (CityGML), archetypal building physics and systems (TABULA, “Regionalised catalogue”), and environmental information for building materials (ÖKOBAUDAT). These sources of information are not limited to the exemplary districts and can thus be principally applied to a large number of districts. While CityGML is the most used, publicly available building model taxonomy in Germany and Europe, it should be noted that building models of CityGML files may be flawed or not compliant with the respective LoD [38], requiring the validation, transformation, or interpolation of such models. This could be achieved using the publicly available toolset DESCity [111]. However, it should be noted that oftentimes CityGML data does not include the buildings’ year of construction, which is essential for the allocation of building archetypes. This necessitates the addition of building-specific data where available. Moreover, the CityGML standard in its newest version 3.0 does not incorporate stages and references to suffice the LCA modelling standards of ISO, EN, or ILCD ([39], Section 1.1.3). The scalability of district-scale LCA using CityGML models could be enhanced if the CityGML modelling format allows for the integration of such LCA information by reducing the effort of creating a common data framework for buildings’ geometries and environmental impacts.

While TABULA covers a broad range of residential buildings, and the regionalised catalogue offers additional information with an even higher spatial resolution (albeit some missing postal code areas and incomplete information for a considerable number of buildings), neither source of information comprises non-residential buildings. The inclusion of such archetypes is crucial to any tool for the large-scale production of ML training and testing data to tackle the current issue of data paucity in the field of building and district-scale LCA (see Section 1.1.4). Analogously, any information on typical building refurbishments should be complemented with statements about their distribution among the respective building type and age class. In addition, there exists no cohesive district typology that could be used for the sake of district-scale building LCA.

ÖKOBAUDAT data is freely available and provides a large number of datasets for the LCI and LCIA of building materials and elements. While both Teco and DisteLCA successfully employ this data, it should be noted that ÖKOBAUDAT’s datasets are not in full compliance with EN 15804 and ILCD. This considers the inconsistent linkage between reference flows and usage of data fields, as well as the occasional lack of impact information for various life cycle modules, potentially violating the respective system boundary definitions (see Figure 2 in Section 1.1.1). This makes it difficult to use ÖKOBAUDAT via its Application Programming Interface (API) and requires additional manual effort for a data framework of CityGML building models, typologies, and environmental information.

A considerable amount of time for the application of both tools for this research has been spent on the preparation of input data from CityGML and ÖKOBAUDAT. A common data framework for CityGML building models, archetypal information on building physics and systems, and related environmental information, could contribute significantly to the efficiency of large-scale LCA applications for buildings and districts. Building and district archetypes for LCA should be dynamic, i.e., reflecting the temporal development of input data besides service lives and considering district effects such as transport.

4.3. The Role of Building Certification Systems and LCA Norms

Building certification systems are a significant contributor to the propagation of LCA on a building and district scale by stipulating detailed statements on energy consumption, material usage, and environmental impacts including $C{O}_2$ emissions throughout the life cycle. This involves the conformity with dataset and methodology standards, and oftentimes a list of valid LCA tools for the scope of the certification (see Section 1.1.3). While district-scale certifications do exist, they currently follow the same granular approach as single-building certifications, hindering the use of heuristic tools such as Teco. Conversely, the widespread building LCA in certifications implies a large production of potential input data for ML applications. However, the natural commercial interest of respective organisations hampers the open-source availability of such data. Nonetheless, providers of building certification could contribute to the propagation of data availability by stipulating a certain degree of spatio-temporal resolution in the LCA-related data.

There is a bandwidth of standards and guidelines for the application of LCA, including the building sector (Section 1.1.1). The aforementioned inconsistency of application and data availability is an ongoing issue (Section 4.2). However, the state of standardisation itself should be critiqued as well. There seems to be a lack of useful guidance on the use of such norms in light of the found inconsistency of data use and LCA norm conformity in published studies. For instance, a more transparent definition of the purported “cut-off criteria” [10,26] would allow for confident use of heuristic tools that explicitly omit certain parts of the scope.

While the use of data generated in the scope of building certification-related LCA in other applications seems questionable, a more heuristic approach to district-scale LCA in such certification systems with a clear focus on comparability could contribute to the widespread use of tools such as Teco. Current standards should be complemented with more coherent guidelines on their use to further enable heuristic estimations.

5. Conclusions

The presented research emphasises the LCA methodology’s fundamental ability to assess the environmental impact of the building sector on a district scale. Ambitious emission reduction objectives (Section 1) and building certification systems (Section 1.1.3) both propel the employment of the method. The normative background as presented in Section 1.1.1 can be translated into tools, receiving necessary input from publicly available data sources. However, both input and output do not currently satisfy the Cradle-to-Gate scope definition. The publicly available data sources discussed and used in this paper show inconsistencies that impede the application’s compliance with current standards, and the overall scalability of the method. Therefore, it is strongly suggested to develop a common data framework to comprise GIS-based building geometry, building physics and systems, and environmental impacts in accordance with current standards, to reduce the amount of manual effort, and to accelerate district-scale LCA of buildings. In this context, a more widespread development and public availability of GIS-based district-scale building LCA tools is desirable for (i) the calibration of the presented heuristics, (ii) a more concise validation of the presented software, and (iii) the general promotion of such workflows and large-scale LCA in the building sector. The further development of the presented or any related tools should incorporate a more detailed consideration of weather and climate, particularly convective heat transfer in the context of wind conditions as well as shading and radiation on opaque surfaces. Such developments have to consider the complexity and runtime of the tool to preserve the time-saving advantage of a heuristic workflow. To tackle the issue of missing archetypes and undeveloped quarter equivalents, the authors propose an archetype development incorporating empirical building data such as energy use, and validation of simulation models. This should involve a dynamic approach, where the temporal development of influential factors on emissions is taken into consideration. Building materials should be identified in a spatial resolution that allows for the identification of regionally distinctive building elements and adherent environmental impacts. Given the minor differences between the analysed districts in terms of weather conditions (negligible differences in outdoor temperature and solar radiation), and locally employed building materials (given data sources do not suggest any significant material difference between the presented locations), the authors suggest extending the presented method onto different regions of Europe. This could be supported by the availability of residential building archetypes for a range of European countries within the TABULA building typology. Such data could potentially be employed in ML for a more convenient determination of LCI and LCIA. In terms of participation, stakeholders such as real estate owners, planners, and inhabitants should be made aware of the necessities and benefits of building LCA, and how they could contribute to the process. Specifically, they can provide detailed information on buildings’ year of construction, geometry, materials, and energy consumption, which can be used as accurate input for the presented methodology. Thus they contribute significantly to the large-scale analysis of a region’s (or country’s) building stock, and sustainability measures such as refurbishment subsidies and other incentives derived from the method’s output data. While the authors greatly appreciate the information supplied by various project partners for the scope of the district analyses, a more generalised process for the handling of such data could simplify its availability. With a broad range of influential factors and the exponentiation of uncertainties, large-scale LCA should rather focus on a common methodology to ensure the comparability between different use case outcomes, rather than a high degree of precision. Future developments of standards should take this into consideration to enable relevant stakeholders to conduct LCA.

The initial research question—“To which extent could a simplified method approximate LCA results to a satisfactory extent while requiring less data input and computational time?” (Section 1)—can be answered such that a simplified method is fundamentally able to output LCA results of a resolution that satisfies the needs of relevant stakeholders, provided that sufficient input data exists. The computational method easily outperforms a manual approach of LCA, albeit the two presented tools show practical differences and overall potential for a further reduction in manual effort. A common data framework and large-scale statistical evaluation of related data can significantly contribute to the large-scale LCA of buildings on a district level for a range of use cases.