District-scale energy and carbon footprint assessments of buildings

The building sector is a major contributor to global and national Greenhouse Gas (GHG) emissions, motivating increasingly demanding regulations and incentives to achieve net zero goals in both retrofit and new construction. This influences the strategic and operative tasks of multiple stakeholder groups in industry, policymaking, and research: i) Urban planning needs to consider more restrictive building codes and elaborate carbon reporting requirements; ii) policymaking seeks to optimise the allocation of emission-reducing retrofit subsidies; iii) European real estate asset management has to consider energy efficiency and emission thresholds for investment decisions as stipulated by the Taxonomy Report’s Technical Annex; iv) research should develop and advance methodologies of emission calculation and impact prediction. Life Cycle Assessment (LCA) has emerged as a standardised method of prediction for material, energy, and emission flows to this end. Moreover, Urban Building Energy Modelling (UBEM) is a subject of increased interest for the context between buildings’ thermophysical properties, energy demands, and environmental impacts. It regularly uses building templates, or archetypes, to enrich and simulate simple geometries in the purported bottom-up approach. However, both LCA and UBEM require large amounts of data, thwarting practical application. Building archetype sets such as Typology Approach for Building Stock Energy Assessment (TABULA) have been developed, yet without embodied GHG emissions or non-residential buildings, and sporadic consideration of regional construction variations. Several UBEM tools include both operational energy and embodied carbon simulations, however, none of the identified programmes adhere to the life cycle stage model of European Standard (Euronorm) (EN) 15804 and EN 15978. Thus, this work intends to fill the laid out research gaps by developing a German building and district typology, using a broad range of data sources, simulation tools, and Machine Learning (ML) methods. This includes the development of the Teco enrichment framework for the determination of buildings’ Global Warming Potential (GWP) according to the stage model, leveraging the district-scale heating energy demand simulation of buildings as Resistance-Capacitance (R-C) models in the Tool for Energy Analysis and Simulation for Efficient Retrofit (TEASER). A parametric simulation run of 81,990 building models containing approximately 200 variables has been used to derive statistical ranges of energy and emission Key Performance Indicators (KPIs). The output was also used to train Gradient Boosted Tree (GBT)-based generative surrogate models for KPI prediction, with an extension of socio-economic and socio-technical variables, or features. The models employ a feature space of 61 to 74 variables. Permutation-based feature importance analysis and Tree-SHapley Additive exPlanation (TreeSHAP) have been performed to analyse the variables’ contributions to the explanatory power of the surrogate models. For each KPI, 40 to 100 % of these variable contributions can be traced back to thermophysical, geometrical, and/or simulative phenomena. For example, the Net Leased Area (NLA) plays a dominant role according to the aforementioned analyses. Likewise, this geometrical variable is essential to the simulation workflow in Teco and TEASER. The developed ML models’ KPI predictions lie within the statistical ranges of the parametric simulation run for 60 to 98 % of the building archetypes. Moreover, the ML models perform one KPI prediction per circa 0.2 milliseconds, whereas the Teco workflow requires between 10 to 300 seconds, depending on the model zone complexity. The aforementioned simulation output has also been used to derive 15 district archetypes, comprising weighted averages of present building templates, for use in a prototypal “District LCA tool” on the website https://hub4lca.e3d.rwth-aachen.de/. The Heuristic Urban Building modelling for Life Cycle Assessments (HUB4LCA) dataset of 977 residential and non-residential archetypes, available as tabular data and scorecards, as well as Teco and the extended generative surrogate model, have been used in demonstrative applications for the aforementioned stakeholders. The assessment of respectively 96, 5,000, and approximately 20 million buildings (the German residential building stock as of 2022) has yielded results that inform operative and strategic decision-making processes. However, the cross-examination of results between Teco and Urban Modeling Interface (UMI) has revealed the need for calibration. Also, Teco’s current limitation to operational heat energy plausibly obscures the major part of some non-residential building types’ energy consumption patterns. In the near future, data access to HUB4LCA could be operationalised with an Application Programming Interface (API); a detailed cross-calibration of UBEM LCA tools would enhance their traceability and informative value; case studies using Geographical Information System (GIS)-based geometry could identify any sensitivities of the GWP assessment to complex cubages; and broader analyses on the influence of socio-economic and socio-technical variables on building energy and environmental performance could foster a top-down modelling approach. In the longer term, a moderate extension of national censuses with three variables, or questions, on buildings’ construction types (facade and roof), estimated insulation thicknesses, and annual heating and electricity consumption, could dramatically increase the availability of datasets for, and accuracy of, LCA in UBEM. The creation of a German national Energy Performance Certificate (EPC) database would have a similar impact.