Data Mining Meets Life Cycle Assessment: Towards Understanding and Quantifying Environmental Impacts of Individual Households

Abstract

To reduce adverse impacts on nature, thus enabling future generations to lead a decent life, deep changes in present human behavior are urgently needed. Policymakers can assume a key role and aim at creating an environment that enables producers and consumers to move towards more sustainable behavioral patterns. However, in order to successfully implement policy interventions and to efficiently invest time and money in the most promising fields of action, policymakers are in need of a highly detailed level of quantitative information on prevailing consumption patterns and production systems. The goal of this dissertation was to investigate and develop new approaches that would be able to provide a thorough information base to support the design, prioritization and implementation of effective environmental policies. Thereby, a special focus was laid on the exploitation of Big Data and the application of data mining and machine learning techniques to extract new information tailored to the respective policymakers’ areas of influence. In order to achieve this goal, a two-track approach was pursued. Initially, building upon in-depth surveys and data collection, a database providing accurate data of local actors and activities was established for the municipality of Zernez, a Swiss mountain village, in the scope of a transdisciplinary research project. Subsequently, a comprehensive spatially resolved modeling framework was developed to estimate similarly detailed data for data-scarce regions. The in-depth analysis of the current carbon footprint in Zernez provided an effective planning basis for the research team to develop a concrete action plan. Yielding a greenhouse gas (GHG) reduction potential of 80%, the building stock was identified as a reasonable first step to devise GHG mitigation strategies in Zernez. The proposed actions could then lead to a reduction of 13% and 17% of the municipality’s total carbon footprint from a consumption and production perspective, respectively. The experiences gained in this project demonstrated the importance of understanding and quantifying the variability of local actors (producers and consumers) to develop and prioritize targeted GHG reduction measures. In order to provide other municipalities with similarly comprehensive information without laborious data collection, an extensive modeling framework was established in a second stage. The models elaborated in this dissertation follow three principles. First, they are built from the bottom-up, which allows for reproducing a realistic picture of the variability of local actors and for aggregating simulation results on any desired regional scale. Second, the models source data from publicly accessible databases and third, they adopt a life cycle consumption-based perspective, which means that they assess resource uses and emissions induced by the consumer demand in a certain area. The overall modeling framework takes thus individual households as central modeling elements and aims at deriving a realistic environmental profile for each household within a certain region. This resolution is important because purchase decisions are made on the level of households. The overall modeling framework encompasses three sub-models: a physically-based building energy model, a data-driven consumption model and a mobility model building upon the results of an agent-based simulation. The overall model was applied to the whole of Switzerland in order to demonstrate its practical feasibility. Still, the concepts of the sub-models are generic enough to be applicable to other countries with comparable data. A global sensitivity analysis was applied to study the building energy model’s internal structure, and the database of Zernez allowed for an in-depth evaluation with primary data. Based on these insights, the model was then improved by integrating comprehensive large-scale geographic data, including the use of nationwide laser-scanning data to derive 3D-building geometries. The final model is able to provide estimates of energy demand for each residential building in Switzerland. A thorough evaluation of the model results with reported data concluded that this model can approximate a realistic picture of the overall characteristics of a certain building stock’s energy demand. The consumption model embarked on a novel approach to assess the variability of lifestyle-induced environmental impacts. Based on an extensive use of data mining techniques, prevailing consumption patterns were studied and 28 consumption-based archetypes derived. These archetypes were further investigated and revealed different behavior patterns within similar socio-economic groups. Furthermore, archetypical behavior deviating from macro-trends, such as increased environmental impacts with higher income, could be detected. The proposed archetype-approach can thus be regarded as a promising basis to foster the understanding of current consumption patterns and to contemplate effective policy measures to reduce consumption-induced environmental impacts. Moreover, these archetypes can serve as building blocks to model the demand of food, services, consumables and other goods of households within a specified region. For this, the archetypes were assigned to real Swiss households by means of a newly developed probability-based classification framework which simultaneously interlinks with both the building energy model and the mobility sub-model. The latter estimates the households’ transport demands based on the results of an agent-based simulation framework that reproduces typical mobility behavior of the Swiss population. The overall model predicts the demands in about 400 different consumption areas for all approximately four million real Swiss households by taking into account the given circumstances of a specific household. A hybrid life cycle assessment (LCA) framework then assesses and subdivides the environmental impacts associated with these demands into more than 200 different categories. The applied LCA allows for computing different environmental indicators, and found that the estimated average consumption-based carbon footprint of Switzerland amounts to 9.5 tons CO2-equivalents per person per year. Besides interesting differences among household consumption archetypes, the large-scale application of the overall model also reveals regional distinctions. For instance, mobility demands in rural areas tend to induce higher GHG emissions than their urban peers. Such differences should be further investigated to identify potential drivers of environmental impacts. The high resolution of the overall model and its ability to quantify the variabilities in both household- and region-specific behavior renders it a powerful information tool to understand locally occurring consumption patterns. It may thus help to find hotspots, identify areas of action, and allow for the designing of impactful environmental measures tailored to specific household groups to reduce their impacts. The model can further be used as a platform to evaluate policy scenarios in upcoming research. The physically- and component-based approach of the building energy model as well as the link to an agent-based mobility model in particular will allow for the analysis of future scenarios in the context of total household consumption. This dissertation demonstrated how Big Data and its analysis techniques can be employed to create a comprehensive knowledge base to inform environmental policymaking at different regional scales. The developed model is a starting point for more detailed investigations and it is open for further developments, improvements and extensions in future.