Advancing Life Cycle Assessment for Emerging Energy Technologies

Thesis submitted in fulfilment of the requirements for the award of the degree of Doctor of Engineering Sciences (Doctor in de Ingenieurswetenschappen) ADVANCING LIFE CYCLE ASSESSMENT FOR EMERGING ENERGY TECHNOLOGIES A MULTI-LEVEL STUDY FOR BELGIUM Dominik Huber October 10, 2025 Promotor: Prof. Dr. ir. Maarten Messagie Co-promotor: Prof. Dr. ir Thierry Coosemans Jury: Prof. Dr. ir. Hubert Rahier, chairman Prof. Dr. Sebastiaan Eeltink, vice-chairman Dr. ir. Maeva Lavigne Philippot, secretary Dr. habil. Cristina Madrid-López Dr. ir. Paula Pérez-López Faculty of Engineering Department of Electrical Engineering and Energy Technology (ETEC) – The Electromobility Research Centre (MOBI) – Innovating the Energy Transition (EVERGI-Team) Dominik Huber © 2025. This publication is licensed under CC BY-NC 4.0. This license enables reusers to distribute, remix, adapt, and build upon the material in any medium or format for noncommercial purposes only, and only so long as attribution is given to the creator. For more information, visit https://creativecommons.org/licenses/by-nc/4.0/ Supervisory and jury committee Promotor: Prof. dr. ir. Maarten Messagie Vrije Universiteit Brussel (VUB), Faculty of Engineering Sciences, Department of Electric Engineering and Energy Technology (ETEC), Electromobility Research Centre (MOBI), Research Group EVERGI. Co-promotor: Prof. dr. ir. Thierry Coosemans Vrije Universiteit Brussel (VUB), Faculty of Engineering Sciences, Department of Electric Engineering and Energy Technology (ETEC), MOBI Electromobility Research Centre, Research Group EVERGI. President: Prof. dr. ir. Hubert Rahier Vrije Universiteit Brussel (VUB), Faculty of Engineering Sciences, Department of Materials and Chemistry, Physical Chemistry and Polymer Science (FYSC). Vice-president: Prof. dr. Sebastiaan Eeltink Vrije Universiteit Brussel (VUB), Faculty of Engineering Sciences, Department of Chemical Engineering. Secretary: Dr. ir. Maeva Lavigne Philippot Vrije Universiteit Brussel (VUB), Faculty of Engineering Sciences, Department of Electric Engineering and Energy Technology (ETEC), Electromobility Research Centre (MOBI), Research Group EVERGI. External Member : Dr. habil. Cristina Madrid-López Center for Energy, Environmental and Technological Research, CIEMAT. Spanish Ministry for Science and Innovation. External Member : Dr. ir. Paula Pérez-López Joint Research Center of the European Commission, Project Officer for Circular Economy and Sustainable Industry (JRC.B.5). i ii Declaration on the Use of Generative AI I acknowledge the use of generative artificial intelligence (AI) tools, specifically ChatGPT developed by OpenAI, during the preparation of this thesis. These tools were employed primarily for language-related assistance, including paraphrasing, grammar improvement, and clarity checks. Additionally, generative AI was used to support the drafting of summary sections, the development of code for LCA models, and the design of data visualizations. All results, numerical data, images, and textual content presented in this thesis are the outcome of my own research and analysis, unless explicitly stated otherwise. The use of generative AI was conducted in accordance with the current guidelines of the Vrije Universiteit Brussel. All AI-generated content was critically reviewed and validated before being incorporated into the thesis. I take full responsibility for the originality, accuracy, and integrity of the work presented herein. iii iv Acknowledgement The true achievement of a PhD thesis is not the research itself. The real accomplishment lies in the dedication to finishing it. Many people told me this over the course of my PhD journey, and I didn’t want to believe it, no matter how many times I had heard it. The fact that you are reading this today means: I actually did finish it. The past five years have been incredibly exciting, filled with highlights, from attending my first conference during COVID, to a research visit in Barcelona, to countless unforgettable, inspiring, and insightful PhD courses. No matter what I dreamed of pursuing, my supervisor and co-supervisors always fully supported me. In fact, it was during an hour-long, engaging, and stimulating online call in February 2020 that I knew I wanted to start my PhD journey with Maarten. I would like to take this opportunity to thank you for your trust, the freedom and flexibility you gave me, and your support throughout my PhD. I was able to build the perfect ecosystem to maximize my personal growth, and I am extremely grateful for that — thank you! This thesis is a good representation of all the things I’ve learned, things I never even dared to dream of five years ago. Of course, it takes more than just supervision and attending a few PhD courses. I was lucky enough to work with a very inspiring and extremely supportive post-doctoral researchers: Maeva. I don’t even know where to start thanking you. In our bi-weekly update calls, you had to listen to my countless complaints about missing data, broken code, incomplete evaluations, etc. No matter how desperate the situation seemed, Maeva always had a solution. Even outside our scheduled meetings, I could always reach you and ask for an emergency call. But Maeva is not just a problem solver; she is also an inspiring researcher. Whether it’s LCA, batteries, social LCA, uncertainty evaluation, or impact categories, Maeva always has an idea or at least a paper at hand where we could find inspiration. She is also a role model when it comes to project work. During her own PhD, she contributed to more than ten projects. I’ve been involved in eight projects during mine. And when things got overwhelming, Maeva showed me how to stay calm and turn anxiety into deliverables. Thank you for your continuous support and motivation! It has been, and continues to be, a pleasure working with you! Another word of gratitude goes to my LCA coworkers Anne, Lea, and Leon. Methodological (sometimes more philosophical) discussions, our weekly team exchanges, attending conferences, and tackling LCA problems have brought us closer together over the years. At this point, I also want to thank all my other EVERGi colleagues for building a social network that was undoubtedly one of the reasons I stuck with this PhD! In particular, I would like to express my gratitude to Ander, Alex, Gilles, and Julian: No matter how many times I texted, called, and bothered you with stupid coding questions, basic engineering issues, GitHub troubles, or other seemingly simple problems. You were always open and helpful, no matter how silly my questions were. I would have needed double the time to complete my PhD without your support. I’m glad that, over time, our professional relationships have turned into friendships. You all have a bright and promising future ahead and I’m excited to see where life will take you! v vi A final word of thanks goes to Cristina Madrid-Lopez. During the autumn school in Switzerland, you invited me to visit your university. Without a long and formal application procedure, I had the opportunity to spend three months at your institute. I not only expanded my perspective beyond engineering and LCA but also gained insights into the structure, work culture, and organization of another research institute. Gracias por esta oportunidad! Beyond my professional network, I also benefited greatly from the social support of my new Brussels family. No problem was too big that it couldn’t be solved over an Apéro, some pizzas, or a good brunch. Thank you, Olga, Pat, Stefano, Federico, Rita, Francesco, Basti, Sandra, Leo, and Audrey. A special thank-you also goes to all my beach volleyball friends, especially Olga & Stefano: No matter how busy or stressed I was, you always managed to distract me with new physical challenges. Over the past few years, beach volleyball became my favorite way to clear my head and focus entirely on the game. This thank-you also extends to all the other Beachers who teamed up with me for fun and challenging matches! Und nun zu meinem wichtigsten Begleiter: Simone. In den letzten dreieinhalb Jahren hast du dir all meine Probleme angehört, Ratschläge gegeben und mich aufgemuntert, wenn es „mal wieder nicht so lief, wie ich mir das vorgestellt habe“. Besonders im letzten Jahr vor der Abgabe, das mich persönlich an meine Grenzen gebracht hat, warst du für mich da. Du hast meine Tiefs etwas seichter gemacht und die Höhen mit mir geteilt. Vor allem aber hast du in aussichtslos erscheinenden Situationen einen klaren Kopf bewahrt und mich mit deinen Tipps und Ratschlägen wieder „back on track“ gebracht. Grazie di cuore! Ti amo! Mein letzter Dank gilt meiner Familie im Unterallgäu, allen voran meinen Eltern: Ihr habt mich zu dem Mann erzogen, der heute hier steht und seine Doktorarbeit verteidigt hat. Ohne eure Unterstützung, euren Zuspruch und eure Liebe wäre ich nie hier angelangt! Diese Thesis ist nicht nur meine Thesis, sondern indirekt auch eure! Ich hoffe, ihr seid genauso stolz auf dieses Buch und auf mich, wie ich es bin. Danke für alles – ich hab euch lieb! Of course, this is not a complete list. To everyone I may have forgotten to mention: thank you—your support, in ways big or small, made this journey smoother and more enjoyable. My heartfelt thanks, Dominik Abstract The European Union aims to decarbonise its energy system by replacing fossil fuels with renewable energy technologies (RET). Among these, wind energy stands out for its low cost and climate change (CC) impact, particularly with innovations enabling turbines up to 15 MW. However, the intermittent nature of RET calls for flexible solutions to balance supply and demand. Electric vehicles (EVs), beyond mobility, offer flexibility by storing and discharging electricity, acting as mobile batteries. Once no longer suitable for propulsion, EV batteries can be repurposed as stationary second-life batteries (SLBs). Life cycle assessment (LCA) is a useful tool to evaluate the environmental performance of these emerging energy technologies. While previous LCA studies explore onshore and offshore wind technologies and second-life batteries, they are often location-specific or lack forward-looking modeling. A pan-European model that incorporates regional differences in sea depth, transport, and infrastructure is still missing. SLB assessments rarely account for evolving background systems or their long-term impacts. Similarly, existing LCA studies on EV flexibility are limited to either a system-level view or static individual use cases, without capturing scale-specific impacts or modeling uncertainties in charging strategies. In light of Belgium’s 11.3 % RET gap under its final National Energy and Climate Plan (2024) [1], this thesis aims to support pathways for Belgium toward the EU’s 33 % renewable energy target by 2030. It introduces improved LCA methodologies focusing on: •Geospatial refinement: Enhancing wind turbine LCA models with location-specific data, floating foundations, and site-specific production to enable system-wide analysis across Europe. •Short-term temporal refinement: A comparative framework combining multi-energy system modeling and LCA, including re-calculated hourly Belgian grid impact data to assess different EV charging strategies. •Long-term temporal refinement:A prospective LCA (PLCA) of SLBs based on the occurrence of life cycle processes and electricity mix projections to 2050. •Multi-scale integration: Applying models at micro (household), mezzo (industrial), and macro (national energy system) scales. This includes integration the TIMES energy system model for evaluating future PV–BESS–EV systems. The thesis also expands beyond deterministic environmental impact analysis, incorporating other impacts than CC, their uncertainty through Monte Carlo and perturbation analysis. It introduces a discernibility assessment and a self-sufficiency ratio for EV flexibility, while wind fleet assessments are enhanced using vii xiv CONTENTS 2.4.3 Assessed impact categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29 2.5 Summary of existing literature and research gaps . . . . . . . . . . . . . . . . . . . . . 29 2.5.1 Advanced life cycle inventory data . . . . . . . . . . . . . . . . . . . . . . . . . 29 2.5.2 Differences in scales when assessing flexibility services . . . . . . . . . . . . . 30 2.5.3 Claimfornewmodels ............................... 31 3 Methodology for assessing emerging energy technologies 33 3.1 Overallmethodology .................................... 33 3.2 Refined geographical inventories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 3.2.1 Goal and scope definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36 3.2.2 Geographically refined inventories . . . . . . . . . . . . . . . . . . . . . . . . . 37 3.3 Refined temporal inventories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 3.3.1 Goal and scope definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40 3.3.2 Temporally refined inventories . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 3.4 Inventories for small and medium scales . . . . . . . . . . . . . . . . . . . . . . . . . . 45 3.4.1 Goal and scope definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45 3.4.2 LCI data collection of small and medium scales . . . . . . . . . . . . . . . . . . 48 3.4.3 Calculations .................................... 51 3.5 Inventories for large scales . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 3.5.1 Goal and Scope Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54 3.5.2 Inventories for Large-Scale Assessment . . . . . . . . . . . . . . . . . . . . . . 55 3.6 Applied life cycle impact assessment methods . . . . . . . . . . . . . . . . . . . . . . . 60 3.7 Uncertainty and sensitivity analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 3.7.1 Background Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64 3.7.2 Foreground Uncertainty and Sensitivity Analysis . . . . . . . . . . . . . . . . . 65 4 Unveiling wind power’s environmental footprint 69 4.1 Usecasedescription .................................... 71 4.2 Individual turbine evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 4.3 Uncertaintyevaluation ................................... 76 4.3.1 Global uncertainty assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 4.3.2 Sensitivity assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 4.4 Macro-levelevaluation ................................... 79 4.4.1 Correlationanalysis ................................ 81 4.4.2 Ordinary least squares regression model . . . . . . . . . . . . . . . . . . . . . . 82 4.4.3 Randomforestmodel................................ 83 4.5 Discussion and limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 5 Assessing flexibility solutions in energy systems 89 5.1 Usecasedescription .................................... 91 5.1.1 Micro-levelusecase ................................ 91 5.1.2 Mezzo-levelusecase................................ 91 5.1.3 Results of improved datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 5.2 Micro-levelresults ..................................... 95 5.2.1 Contributionanalysis................................ 95 5.2.2 Uncertainty evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 5.2.3 Other impact categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 5.3 Mezzo-levelresults..................................... 101 5.3.1 Contribution analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 CONTENTS xv 5.3.2 Uncertainty evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 5.3.3 Other impact categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 5.4 Discussion and limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107 6 Unlocking the potential of second-life batteries in energy systems 109 6.1 Usecasedescription .................................... 111 6.1.1 Firstlife....................................... 111 6.1.2 Usecases...................................... 111 6.2 Results............................................ 112 6.2.1 Residentialusecase ................................ 114 6.2.2 Industrialusecase ................................. 114 6.2.3 Utilityusecase................................... 114 6.2.4 Effects of the Belgian Pathways on SLB Impacts . . . . . . . . . . . . . . . . . 115 6.2.5 Other environmental impacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117 6.3 Discussion and limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 7 Decentralized roll-out of photovoltaic systems 121 7.1 Energybalances....................................... 123 7.2 Environmentalresults.................................... 125 7.2.1 Positive implications on the environment . . . . . . . . . . . . . . . . . . . . . 125 7.2.2 Negative implications on the environment . . . . . . . . . . . . . . . . . . . . . 127 7.3 Sensitivity.......................................... 129 7.4 Discussion and limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 8 Conclusion, recommendations, and outlook 133 8.1 Contextualization...................................... 134 8.1.1 Implications for the Belgian energy transition . . . . . . . . . . . . . . . . . . . 134 8.1.2 Improved LCA methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 8.1.3 Lessons learned from the applied research design: constructive reflections . . . . 137 8.2 Closingthegap ....................................... 138 8.3 Stakeholder-specific recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 8.3.1 Electricity consumers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 8.3.2 Industry....................................... 141 8.3.3 National policy makers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142 8.3.4 Scientificcommunity................................ 142 8.3.5 Broadersociety................................... 143 8.4 Futureoutlook........................................ 143 A Overview of academic activities and research contributions 145 B Further information about the literature reviews 151 C Additional visualizations of results 159 C.0.1 Unveiling Wind Power’s Environmental Footprint . . . . . . . . . . . . . . . . . 159 C.0.2 Assessing Flexibility Solutions in Energy Systems . . . . . . . . . . . . . . . . 174 C.0.3 Micro-levelresults ................................. 176 C.0.4 Mezzo-levelresults................................. 180 C.0.5 Unlocking the Potential of Second-Life Batteries in Energy Systems . . . . . . . 187 C.0.6 Decentralized roll-out of Photovoltaic Systems . . . . . . . . . . . . . . . . . . 196 Chapter 1 Introduction 1.1 The Belgian energy transition The European Union (EU) has committed to achieving climate neutrality by 2050 through a comprehensive set of legislative and policy initiatives. Central to this ambition are the European Green Deal, the Fit-for-55 package, and the REPowerEU plan. While the Green Deal outlines the overarching goal of net-zero emissions by 2050 [2], the Fit-for-55 package sets a binding target to reduce greenhouse gas (GHG) emissions by at least 55 % by 2030 [3]. The REPowerEU plan, adopted in response to the geopolitical energy crisis, aims to reduce dependency on Russian fossil fuels and accelerate the deployment of renewable energy [2–4]. In this context, life cycle assessment (LCA) plays a pivotal role to present science-based and robust results to assess strategies for decarbonization. To ensure that these objectives are effectively translated into national action, Regulation (EU) 2018/1999 on the Governance of the Energy Union and Climate Action requires each Member State to develop a National Energy and Climate Plan (NECP) [5]. These plans must outline concrete policies and measures across five dimensions: decarbonization, energy efficiency, energy security, the internal energy market, and research and innovation. Belgium’s NECP sets a target of 21.7 % renewable energy in gross final consumption by 2030 [1]. To approximate this share, Equation 1.1 is applied [6]: RET =PRET Cf (1.1) Where: •RET : Share of electricity produced from renewable energy technologies (RET) (%), •PRET : Primary production of electricity from renewable energy technologies (kWh), •Cf: Gross final energy consumption of Belgium including the industry, transport, households, services incl. public service, agriculture, forestry and fishery, the consumption of electricity and heat by the energy branch for electricity and heat production and transmission and distribution losses (kWh). In 2023, Belgium emitted approximately 98,123 ktCO2eq, with transport (25 %), the energy industry (16 %), the processing industry (16 %), and the combustion sector (12 %) as the main contributors [7]. A key decarbonization strategy for the transport sector is the electrification of the vehicle fleet. However, 1 2CHAPTER 1. INTRODUCTION the full climate benefit of electric vehicles (EVs) can only be realized if the electricity used to charge them is also decarbonized. This interdependence places additional pressure on the energy sector to transition towards low-carbon electricity sources. Once a high share of RET has been introduced to the national energy system, transport electrification will approach saturation. Renewable energy technologies, particularly wind and solar power, are among the most promising solutions for decarbonizing electricity generation. In addition to their low levelised cost of energy (LCOE), renewable sources generally impose fewer external environmental and health costs over their life cycles. As shown in Figure 1.1, wind energy, especially offshore wind, exhibits the lowest external costs compared to fossil-based electricity generation [8]. External costs refer to indirect burdens not reflected in market prices, such as environmental degradation or health impacts. Figure 1.1: Life-cycle external costs of electricity production from renewable and fossil energy sources. Blue lines indicate the range of climate-related externalities, and red lines reflect air pollution health impacts. External costs are higher for fossil energy unless equipped with carbon capture and storage (CCS) (Comb.C: Combined cycle; Postcom: Post-combustion; η: efficiency factor; PV = Photovoltaic; source: [8]). Conversely, renewable energy technologies may also generate external benefits, such as social and economic advantages like as job creation or improved energy resilience [8]. Offshore wind energy is particularly favorable in terms of both climate and health-related externalities. Photovoltaic (PV) systems, while advantageous compared to fossil fuels, show comparatively higher external costs within the spectrum of renewable options, particularly when deployed in less favorable locations. Solar energy, however, offers key advantages for deployment. With minimal permitting requirements, limited infrastructure needs, and competitive investment costs (approximately 850 EUR/kWp for residential PV installations in Belgium [9]), PV installations can be implemented quickly and at distributed scales. In contrast, wind energy, despite its lower external costs, faces challenges due to lengthy and complex planning and permitting procedures [10]. 1.2. RENEWABLE ENERGY DEPLOYMENT IN BELGIUM 3 The increasing penetration of RETs also has broader implications for how energy systems are organized. Traditional systems rely on centralized power generation, where electricity is produced at largescale power plants and then distributed across the grid. RETs, by contrast, enable local generation and operation, thereby fostering a shift toward more decentralized energy systems. Decentralized and centralized energy systems differ across several dimensions, including connectivity, proximity, flexibility, and controllability [11]. In their framework, Bauknecht, Funcke, and Vogel define connectivity as the grid level to which power plants are connected, typically the distribution grid for decentralized systems and the transmission grid for centralized ones. Proximity refers to the spatial distance between electricity generation and consumption. Flexibility relates to the capacity of the system to adapt to changing demand and supply, while controllability concerns the means of balancing generation and consumption in real time [11]. Although the share of renewables has increased steadily, it stood at approximately 15 % in 2023, well below the upcoming 2030 goal, as illustrated in Figure 1.2. Figure 1.2: Gross electricity production in Belgium by source (RET = renewable energy technologies [12]). In parallel with efforts to expand renewables, Belgium has also revised its nuclear energy policy. Initially planning for a complete nuclear phase-out, the government decided in May 2023 to extend the operation of two reactors, Doel 4 and Tihange 3, until 2045, and announced plans for new nuclear investments [13]. This decision reflects a broader reassessment of the role of nuclear energy in ensuring a secure electricity supply during the energy transition. The gradual shift towards a more sustainable electricity supply in Belgium is reflected in both existing infrastructure and ambitious future deployment targets, as outlined in its NECP. 1.2 Renewable energy deployment in Belgium Building on its climate ambitions, Belgium’s NECP sets out concrete targets for renewable energy deployment. By 2030, the country aims to install nearly 12 gigawatts (GW) of wind capacity, including 5.8 GW offshore. This expansion is projected to contribute approximately 34.8% to gross final energy consumption [1]. Additionally, the NECP outlines an increase in solar PV capacity to 8.5 GW by 2030. Looking further ahead, Belgium plans to expand wind capacity to 15.6 GW by 2040, of which up to 7 GW would be offshore [1]. Despite these ambitions, the European Commission notes that Belgium’s NECP remains insufficient to meet the EU-wide target of achieving a 33 % share of energy from renewable sources in gross final 4CHAPTER 1. INTRODUCTION consumption by 2030 [1]. According to the latest data from the European Network of Transmission System Operators for Electricity (ENTSO-E) shown in Figure 1.3, Belgium currently has approximately 8.8 GW of installed solar capacity, 3.1 GW of onshore wind, and 2.3 GW of offshore wind [14]. Meeting the 2030 goals would require additional installations of approximately 2.9 GW of onshore and 3.5 GW of offshore wind capacity. Reaching the 2040 targets would necessitate an additional 2.6 GW onshore and 2.7 GW offshore. A key element of Belgium’s offshore wind strategy is the development of the Princess Elisabeth Zone (PEZ), officially announced in 2021. The PEZ is expected to deliver between 3.15 and 3.5 GW of new offshore wind capacity in the Belgian North Sea [15]. To enable this, three public tenders are scheduled between 2024 and 2028: •Princess Elisabeth tender 1: Installation of 700 megawatts (MW) in an area of 46 km2 •Princess Elisabeth tender 2 and 3: Up to 1,400 MW each in areas of up to 107 km2 The first tender (Princess Elisabeth tender 1) is currently in preparation, with the submission window expected to open on July 24th for a period of one month [15]. These developments highlight both the scale of Belgium’s renewable energy ambitions and the challenges associated with meeting them within the prescribed timelines. To effectively integrate increasing shares of wind and solar power, the internal energy market must adapt to their intermittent nature. Section 1.3 discusses the role of flexibility and connectivity in this evolving system landscape. Figure 1.3: Evolution of installed wind and solar capacity in Belgium [14]. 1.3 Market connectivity and flexibility Another mandatory reporting point of the NECP concerns the internal energy market. Due to their intermittent nature, a high deployment of renewable energy technologies in the national energy systems poses a challenge. Fortunately, technological solutions exist to mitigate the fluctuating nature of renewables. One common solution is to store unused electricity during periods of overproduction and release it when needed. However, stationary storage systems such as battery installations remain expensive, and their production and raw material extraction can lead to additional environmental burdens [16]. As an alternative to installing dedicated storage infrastructure, the existing batteries already integrated into EVs could help store surplus electricity generated from renewable sources. In fact, individual passenger EVs are parked around 95 % of the time [17]. Simultaneously, EV sales have grown steadily over the past decade. These developments suggest a potential synergy between the increasing availability of EV batteries and the rising penetration of intermittent renewable energy technologies. 1.3. MARKET CONNECTIVITY AND FLEXIBILITY 5 With EVs becoming more widespread, and with battery capacities increasing, most vehicles are idle for extended periods, while their electricity demand does not necessarily need to be met immediately upon connection. This opens the possibility to reschedule charging to align with periods of low electricity prices or high renewable energy availability. While EVs are typically charged immediately upon arrival at a charging station (referred to as uni-directional charging), more advanced strategies aim to shift charging in time to support grid integration of renewables. This approach is referred to as smart charging. Smart charging allows for greater flexibility by aligning electricity demand with supply patterns, for example by shifting EV charging to hours with high solar or wind generation. Over the years, additional flexibility services such as peak shaving or grid balancing have also been explored [18–20]. At the same time, those services can help avoid grid congestion, which may occur when renewable sources produce electricity simultaneously, for example on sunny or windy days. In contrast to upgrading grid infrastructure, which is a costly and long-term undertaking, using EV batteries as mobile storage through smart charging has emerged as a promising alternative for accommodating more renewable energy. However, these go beyond the scope of this thesis, which focuses on comparing uni-directional and smart charging strategies. Nevertheless, to effectively utilize EV batteries as a source of flexibility, improvements in market connectivity and digital infrastructure are necessary. This includes empowering final consumers, deploying smart meters, and enabling demand-side management mechanisms. While the European Commission acknowledges Belgium’s progress in enhancing interconnection within the internal energy market, it has also criticized the updated NECP for lacking concrete measures to accelerate the deployment of electricity storage and for insufficient engagement of system operators in promoting flexibility services [1]. Even though providing flexibility services during their operational life could promise potential benefits, criticism arises regarding the sustainability of EV batteries. Apart from environmental and human rights violations due to unsustainable extraction of materials, EVs are criticized for having their batteries retired at stages where they are still fully functional but do not match the demanding performance required for driving. Under normal conditions, EV batteries are considered to reach their end-of-life (EoL) once their remaining capacity, defined as the state of health (SoH), diminishes to around 80 % or lower [21]. Extensive electricity demand for powering the EV requires a minimum SoH of around 80 %. Current practice foresees recycling the retired traction batteries once they have reached their EoL criterion. However, criticism arises as those batteries are recycled even though they are still fully functional, though unfit for intensive application in EVs. Recognizing the economic value of retired EV batteries and their potential to reduce climate change (CC) impacts, various alternatives to recycling are getting more and more attention. One concept, repurposing, foresees giving retired EV batteries, partly or completely, a second life in a different application as in its first life, such as stationary energy storage [22]. In this process, retired EV batteries undergo tests and refurbishment to suit diverse secondary applications. Repurposing retired EV batteries would allow better utilization of the batteries and thus could be a compelling proposition for stakeholders across various sectors. At the same time, repurposing retired EV batteries could reduce costs and CC impact, as it would avoid manufacturing a new battery for the second-life application. Due to its market structure, Belgium serves as a country of particular interest for the application of second-life batteries (SLB): With 34 % of newly registered vehicles in 2022 being hybrid or full EV and the fact that over 60 % of the newly registered vehicles are company cars, Belgium is foreseen to face rapid transportation electrification, opening a large market for SLB [23]. These strategies, ranging from flexibility during first-life use to second-life battery applications, interact differently depending on the scale of the system under consideration. In this thesis, three system levels are distinguished to capture these varying dynamics: the micro-level, referring to small-scale systems such as individual households; the mezzo-level, representing medium-scale systems such as commercial buildings, industrial sites, or business parks; and the macro-level, encompassing large-scale or national energy systems. The prefix micro originates from the Greek word mikros, meaning “small”, and 6CHAPTER 1. INTRODUCTION refers to phenomena at the smallest or most detailed level. Conversely, macro derives from the Greek word makros, meaning “large”, and denotes phenomena at a broader or systemic level. In economics, these terms are well established: microeconomics examines the behavior and decision-making of individual agents such as households and firms, whereas macroeconomics addresses aggregate outcomes at the scale of entire economies, including growth, inflation, and employment. In contrast, mezzo stems from the Latin medius (“middle”) and designates an intermediate scale between micro and macro. Within the scope of this thesis, the terms micro, mezzo, and macro are employed to differentiate between energy system sizes. Energetic definitions of these different systems used in the thesis are given in Section 3.4.2 and Section 3.5 and their electricity generation is provided in Section 5.1 and Section 7.1. This typology supports the assessment of environmental impacts across different contexts of deployment, technological integration, and policy relevance. In the context of wind turbines, the concepts of micro and macro level are also applied: The different scales refer to either single turbines (micro) or a group of turbines, clustered as a fleet or farm (macro). The previous Sections outlined the overarching policy targets and technological pathways shaping the Belgian energy transition. With increasing shares of variable renewable energy and a growing fleet of electric vehicles, Belgium faces both opportunities and challenges in building a low-carbon, flexible energy system. Flexibility strategies such as smart charging can facilitate renewable integration, while second-life applications for EV batteries offer an additional pathway to enhance sustainability by extending battery use. However, the real-world environmental performance of these strategies remains uncertain and highly context-dependent. Accurately capturing these dynamics, both in space and time, requires robust environmental assessment methods. The following Sections present the methodological and scientific challenges associated with conducting life cycle assessments (LCA) of emerging technologies in a prospective, national context. 1.4 Challenges While LCA is well established for conventional technologies, its application to emerging energy systems in a prospective, national context remains underdeveloped. In general, LCA is a standardized method to quantify environmental impacts of products or services. Conducting an LCA is divided into four phases: goal and scope definition, life cycle inventory analysis, life cycle impact assessment, and interpretation [24]. Traditionally, LCA studies evaluate existing products or systems, thereby representing technologies and market conditions from the recent past relative to the time of study. Such studies are referred to as retrospective LCAs. In contrast, an increasing number of studies aim to model products or services at a future point in time, referred to as prospective LCAs (PLCAs). In a prospective context, current technologies at the laboratory or pilot scale—such as SLB tested and repurposed in controlled environments—are modeled to represent their potential large-scale deployment in the future. A major challenge in PLCA lies in upscaling these early-stage technologies from lab scale to industrial production or market introduction, which inherently introduces uncertainty. Beyond upscaling, a crucial aspect of PLCA involves the definition and development of scenarios that capture possible future developments, such as projected changes in background systems (e.g., the electricity mix or material supply chains) over time. [25, 26]. Future impacts may decrease as a result of learning effects during the upscaling of emerging technologies in the foreground system, as well as through progress and technological changes in the broader economy within the background system [27]. Scientific literature on conducting LCA of emerging energy technologies in a country-specific context leaves various research gaps, which are elaborated further in Chapter 2. Those gaps can then be grouped into clusters, each highlighting certain challenges, which are described in more detail in this Section. 1.4. CHALLENGES 7 Advanced temporal life cycle inventory (LCI) data The LCI stage in LCA is time-consuming and might differ in terms of level of detail required compared to other methodologies, e.g. for energy system models. To comprehensively capture environmental impacts of such advancements of emerging energy technologies requires more refined LCI collection. This cluster can be further broken down into various layers, such as a temporal layer, which explicitly accounts for changes over time in processes, technologies, or background systems. In other words, temporal refers to the representation of time-dependent variations in LCI, such as hourly operation profiles, seasonal fluctuations, or long-term technological transitions. A good example of the temporal layer is the assessment of flexibility services or SLB: When combining, for example energy system models, capturing temporal dynamics of power behaviors with a very high time-resolution, the available average data covered in commercial datasets are not capable of appropriately account for such dynamics. Simultaneously, assets where activities occur over a long time period claim for including background changes over time. The Challenges for this cluster are: 1.1 Accounting for life cycle processes and stages occurring at different time periods in the future 1.2 Obtaining region-specific LCI data for setting up a SLB value chain in Flanders, Belgium 1.3 Remodeling the Belgian national electricity mix using PLCA databases Advanced spatial LCI data Next to the temporal dynamics, more geo-spatial dynamics when conducting LCA of emerging technologies such as wind turbines can be incorporated. In fact, particularly the assessment of RET such as wind energy might be affected regionally, not only in the upstream supply chain for manufacturing components, but also to obtain a more refined, location-specific production of the RET. While capturing those geo-spatial dynamics are demonstrated on individual turbine level or at maximum a country level, no model is currently available for evaluating environmental impacts of both onand offshore wind energy. Therefore, the following Challenges can be derived: 2.1 Building an LCA model that can be applied across all Europe for onand offshore wind turbines 2.2 Calculating the annual electricity production for a country-level wind fleet 2.3 Understanding the behavior and drivers of fleet-level impacts Differences in scales Even though within the goal definition of every LCA the DIN ISO 14040 requires a definition of the intended audience, this cluster goes beyond: The assessed emerging technologies can be performed in different contexts, which might offer different environmental benefits for affected stakeholders. Implications of assessing different scales directly affect the definition of the system boundaries, which might be re-defined for every scale. In a combined approach of energy system modeling and LCA, assessing different scales might even require a change of the energy system model. For example, when assessing flexibility services of EVs, the assessment at household level might require using a highly detailed energy system model, whereas assessing the impacts of the same services at a national EV fleet level asks for a very low-detailed energy system model. Identified Challenges of this cluster are: 3.1 Designing a methodology that captures impacts across micro–mezzo levels, including the impacts of EV batteries used for stationary storage 3.2 Collecting data with sufficient temporal resolution to capture evaluated system dynamics 3.3 Revealing whether certain EV charging strategies offer environmental advantages, even under uncertainty 8CHAPTER 1. INTRODUCTION Need for updated models The first two clusters converge on a third: the need for updated LCA models and databases to translate the inventories into environmental impacts. Furthermore, developed LCA models face the requirement to be universally applicable. Thus, on one side, the LCI collection needs to be able to obtain data as detailed as possible, while on the other side it should come along with enough flexibility to be applied in various contexts. Within those clusters, the following Challenges are defined: 4.1 Integrating an energy system optimization model and LCA in a Belgian context 4.2 Combining top-down macro-level assessments with bottom-up micro-level LCAs 4.3 Evaluating future system dynamics in a holistic and forward-looking manner An overview of research gaps, clusters, challenges and where to find further explanation is provided in Table 1.1. Table 1.1: Research gaps (RG), related challenges, and corresponding chapters addressed in this thesis. The Chapter column indicates where further explanations on the RGs can be obtained (PLCA = prospective life cycle assessment; AEP = annual electricity production; ESM = energy system model). Research gap (RG) Clusters Challenges Sect. RG1: Absence of long-term temporal dynamics in LCI modeling, despite considerable, expectable background changes. Advanced temporal LCI data 1.1. Accounting for life cycle processes & stages occurring at different future time periods 1.2. Obtaining region-specific LCI data on setting up a SLB value chain in Flanders (BE) 1.3. Remodeling the Belgian national electricity mix using PLCA databases 2.3 RG2: Lack of geographically generalizable, spatially refined LCI models in environmental assessment of wind energy across Europe. Advanced spatial LCI data 2.1. Building a LCA model applicable across Europe for on- & offshore wind power 2.2. Calculating the AEP on a country-level wind fleet 2.3. Understanding the behavior and drivers of fleet-level impacts 2.1 RG3: Missing multiscale assessment frameworks to compare different flexibility services provided at different scales. Differences in scales 3.1. Designing a methodology that captures impacts across micro–mezzo levels, incl. impacts of EV batteries for stationary storage 3.2. Collecting data with sufficient temporal resolution to evaluate system dynamics 3.3. Revealing whether certain EV charging strategies offer environmental advantages, even under uncertainty 2.2 RG4: No existing demonstration of the evolution of environmental impacts from a national roll-out of renewable energy systems supported by stationary storage considering different flexibility criteria in Belgium. Claim for updated models 4.1. Integrating ESM & LCA for BE 4.2. Combining top-down macro assessments with bottom-up micro LCAs 4.3. Evaluating future system dynamics in a holistic and forward-looking manner 2.2 Chapter 2 Literature on emerging energy technologies The objective of this Chapter is to present the state-of-the-art scientific literature relevant for the identified topics. Thereby, the literature review should allow the identification of relevant research gaps in the LCA of emerging energy technologies. In terms of methodology, a screening of scientific databases is conducted for all topics. Only for LCA of wind energy a systematic review is conducted, which will be further explained in Section 2.1.1. The state-of-the-art literature is presented either thematically or per use case. In addition to the textual and tabular descriptions in the main body, further information is provided in Appendix B. The last Section 2.5 of this Chapter summarizes the main findings of the review and derives 4 Research Gaps. 2.1 LCAs of wind energy Against the backdrop of steadily decreasing levelized cost of electricity, wind power has become one of the most competitive energy sources of the 21st century [44]. Beyond its cost-competitiveness, wind energy is characterized by low environmental impacts [45]. While no fuel combustion occurs during operation, environmental burdens still arise from material extraction, component production, and infrastructure development. However, the overall environmental impacts over a wind turbine’s lifetime remain lower than those of any fossil fuel technology. Despite these advantages, some critics argue that regional conditions must be considered to fully assess the environmental impact of wind power. Additionally, commercial databases are lacking technological advancements, e.g. they only include the wind turbine construction up until 4.5 MW (based on ecoinvent 3.9.1 database) [46] and represent market datasets reflecting country average technologies, neglecting geographical differences. Lately, Vestas installed a 15 MW turbine in Denmark [47]. To identify relevant studies focusing on LCAs of wind energy which capture those geographical differences, a systematic literature review is conducted and is further presented in the subsequent section. 2.1.1 Systematic Review Approach A systematic literature review is conducted to identify studies that commissioned LCAs of wind energy. To reveal relevant literature, the search terms and the databases are defined. The following search terms are applied: 15 16 CHAPTER 2. LITERATURE ON EMERGING ENERGY TECHNOLOGIES • Primary search string: “wind” • Secondary search string: ”Life cycle thinking”, ”sustainability”, ”life cycle assessment”, ”LCA” Only peer-reviewed, scientific articles are defined as relevant in this literature review, and thus, the targeted database is Elsevier’s ScienceDirect. Peer-reviewed, scientific articles are targeted due to their increased transparency regarding LCI data. The literature screening is performed with a Python script based on the Python package ‘pyscopus’ [48,49]. It helps to search Elsevier’s database for papers containing the above mentioned terms. For relevant identified studies, descriptive information, such as titles, digital object identifiers (DOIs), journals, abstracts, etc., are exported into a spreadsheet. A similar approach in a different context is followed in Bekirsky et al. (2022) [50]. The outcome of the initial search lists 1,194 unique papers. Next, two qualitative filtering processes are undertaken: first, papers are deemed relevant if their abstract relates to wind energy. This first filtering reduced the number of relevant studies to 182. In a following step, those studies are all reviewed with the aim to understand whether they include any regional modeling. All types of regionalization are included, regardless of whether they target the national level, entire wind farms or only single turbines. As a result, 19 out of 182 studies are deemed relevant and are further presented in the subsequent sections, which are divided as follows: the first section gives an overview on studies with a simplified regionalization approach. The second section describes recent advancements of regionalized modeling on an individual turbine level, whereas the third section summarizes latest advancements combining regionalized modeling at larger scale. Although this review follows a structured and transparent process, no formal systematic review protocol (such as PRISMA) is applied. Instead, the methodology is tailored to the scope and needs of this research, prioritizing relevance and replicability over strict adherence to protocol guidelines. This flexible approach allows a targeted focus on the intersection of life cycle assessment and geographical differentiation in wind energy systems. Furthermore, in the scope of the WIMBY project, correlations between materials per component and rated capacity are explored. In addition, in the deliverable, semi-structured interviews with stakeholders involved in the EoL treatment of wind turbines are conducted and it includes findings in a qualitative assessment [34]. 2.1.2 A first approach to regionalization of LCA of wind energy At the macro level, existing literature explores how geographical differences influence the life cycle impacts of wind turbines at a country scale. One approach to regionalize environmental impacts is to account for location-specific electricity yields. Since wind speeds vary by location, annual electricity generation differs across sites, directly influencing environmental results when impacts are expressed per kilowatt hour (kWh) of electricity produced. A considerable number of studies focus on the Chinese wind fleet, both onshore ([45, 51–54]) and offshore ([55]), with 8 out of 19 reviewed studies centered on China. Xu et al. (2022) developed simplified LCA models for the Chinese onshore wind fleet to calculate region-specific CC impacts, incorporating local capacity factors (CFs) based on wind conditions. They reported an average CC impact of 19.88 gCO2eq/kWh [51]. More recently, Li et al. (2023) conducted a similar study for the Chinese onshore fleet, reporting a slightly higher CC impact of 24.9 gCO2eq/kWh [52]. Like Xu et al. (2022), Li et al. (2023) used electricity yield data from local wind speeds to regionalize environmental impacts. Li et al. (2021) extended the scope by incorporating transportation and infrastructure into the LCA of a 40 MW onshore wind farm in China, determining a maximum CC impact of 28.2 gCO2eq/kWh [53]. Building upon these regional assessments, Wang et al. (2021) scaled up the analysis to the entire Chinese power sector, comparing wind power with other energy sources. Using historical data, they assessed emissions of CO2, NOx, and SO2, integrating local power generation into their approach [54]. 2.1. LCAS OF WIND ENERGY 17 While the aforementioned studies focused on onshore wind, Yang et al. (2018) applied LCA to a Chinese offshore wind farm, estimating an average CC impact of 25.5 gCO2eq/kWh [55]. Additionally, two studies explored the future environmental impacts of wind energy, though these are not detailed here [45,56]. Collectively, these studies demonstrate that regionalizing wind power LCAs by incorporating electricity yield variations impacts the environmental impacts. The findings indicate that CC impacts for onshore wind range from 19.88 to 28.2 gCO2eq/kWh, while offshore wind is found to have 25.5 gCO2eq/kWh CC impact. Without regionalized LCI data, average CC impacts found in literature of offshore turbines are 2.0 gCO2eq/kWh, while for onshore wind turbines an average CC impact of 9.7 gCO2eq/kWh is identified [34]. However, these spatial assessments primarily focus on regional electricity production, leaving room for further refinement. Moving on from the macro-level perspective, the following section provides an overview of how LCAs of individual wind turbines or smaller wind parks could be enhanced by integrating additional regional factors, such as transportation or infrastructure impacts, to provide a more comprehensive understanding of wind power’s environmental footprint. 2.1.3 Regionalized LCAs of wind energy at the micro-level Still other ways exist to integrate regionalized data into LCAs at a micro level. Morini, Hotza, and Ribeiro (2022) considered transportation distances for a 1.5 MW onshore wind turbine in Portugal and found that CC impacts could be as low as 4.77 gCO2eq/kWh [57]. Other studies have followed similar approaches [58]. Expanding on this, Gao et al. (2019) investigated not only the effect of transportation distances on wind turbine environmental impacts but also different locations of electricity production. Their study revealed that the CC impact of wind turbines installed in the Gobi desert, a region between northern China and southern Mongolia, could be as high as 51.5 gCO2eq/kWh [59]. Beyond transportation and production location, additional studies incorporate further regional factors into LCAs. Reimers, Özdirik, and Kaltschmitt (2014) assessed a 5 MW offshore wind farm in the German North Sea, incorporating shore distance and sea depth as LCA parameters. These parameters influenced the size of the foundation (jacket structure), the length of the grid connection cables, and transit requirements for maintenance [60]. Their results indicated a CC impact of 16.8 gCO2eq/kWh and highlighted that while parameters related to annual electricity production had a strong influence on GHG emissions, factors such as shore distance and water depth had a more limited impact [60]. Taking a more geo-spatially refined approach, Huang et al. (2022) developed spatially detailed LCI models for Chinese offshore wind turbines, incorporating water depth, air density, wind speed, transmission distances, and transportation distances [61]. For a medium scenario featuring a 10 MW turbine with a 25-year lifetime and a monopile foundation, they calculated CC impacts ranging from 4.6 to 19 gCO2eq/kWh, with deep-water installations associated with higher CC impacts [61]. Shifting the geographical focus to Europe, Padey et al. (2012) developed a simplified LCA model to assess the CC impacts of wind turbines, considering three main factors: technological aspects (e.g., turbine model), geographical influences (e.g., regional wind speeds), and LCA methodology (e.g., turbine lifetime). Their analysis of 17 installed wind turbines in Europe highlighted wind speed and turbine lifetime as the most sensitive parameters [62]. They reported CC impacts for turbines with a lifetime between 10 and 30 years, where turbines operating at low wind speeds (up to 6.5 m/s) had CC impacts between 8.7 and 76.7 gCO2eq/kWh, while those with wind speeds above 6.5 m/s ranged between 4.5 and 22.2 gCO2eq/kWh [62]. In a follow-up study, Padey et al. (2013) applied global sensitivity analysis to identify key LCA parameters, including foundation, tower, rotor, nacelle mass, load factor, nominal power, lifetime, availability, and material composition. Based on these findings, they constructed a simplified LCA model to evaluate wind turbine environmental impacts [63]. In contrast to the studies presented in Section 2.1.2, which regionalize environmental impacts through local electricity production, the studies discussed here integrate regional factors directly into the LCI 18 CHAPTER 2. LITERATURE ON EMERGING ENERGY TECHNOLOGIES phase rather than only normalizing overall impacts. The primary focus of these works is on incorporating transportation distances, shore distance, sea depth, and local electricity production. While they provide valuable insights into the impact of these regional parameters at an individual turbine level, their methodologies have yet to be extended to fleet-wide assessments. 2.1.4 Advancing spatially refined LCA: A macro-level perspective Unlike studies focusing on individual wind turbines, Tsai et al. (2016) analyzed a wind farm consisting of 100 Vestas V112-3.0MW turbines, integrating the effects of location factors, water depth, and distance to shore [64]. They found that the turbine with the lowest CC impact (25.6 gCO2eq/kWh) was located 5 km from shore, despite only having an average energy yield [64]. While this study effectively demonstrates how LCI data can be regionalized at the wind farm level, it is limited to offshore turbines in Michigan’s Great Lakes. Similar efforts have been made for European offshore wind farms. Poujol et al. (2020) conducted a geo-located LCA of a floating offshore wind farm in France, consisting of four 6 MW turbines with semi-submersible floating foundations [65]. To refine the spatial resolution, they incorporated locationspecific data on distance to shore, transportation, and electricity yield. Their results indicate a CC impact of 22.3 gCO2eq/kWh. This study provides a comprehensive spatially explicit LCA for offshore wind in Europe. However, like Tsai et al. (2016), its scope is limited to France and a single foundation type. A broader, more comprehensive spatially refined LCA was conducted by Sacchi et al. (2018) and later expanded by Besseau et al. (2019) [66,67]. Sacchi et al. (2018) developed a parameterized LCA model for Denmark, incorporating technological, temporal, and geographical LCI data [66]. Beyond developing scaling models based on technical parameters, they integrated geographic data to link turbines to the national grid, considered different material supply markets, and accounted for temporal changes in steel and electricity datasets. Reported CC impacts for onshore turbines (100 kW, 500 kW, 1,000 kW, and 2,000 kW) and offshore turbines (2000 kW) ranged from 45.3 to 11.7 gCO2eq/kWh [66]. Building on this work, Besseau et al. (2019) introduced a temporal dimension by assessing past, present, and future environmental impacts of the Danish wind fleet, developing an online interactive platform [67]. Their analysis incorporated changes in the electricity mix and the recycled steel content. They observed a steep reduction in Denmark’s wind fleet emissions, from 40 gCO2eq/kWh in 1980 to 13 gCO2eq/kWh projected for 2030 [67]. More recently, Feng and Zhang (2023) analyzed the environmental impacts of Chinese onshore and offshore wind turbines [68]. Their study examined the influence of geographical location, turbine technology, and operational management, incorporating wind speeds, foundation types, and transportation and maintenance distances. They reported CC impacts of 5.84–16.71 gCO2eq/kWh for onshore turbines and 13.30–29.45 gCO2eq/kWh for offshore turbines, concluding that geographical factors and turbine technology had only a marginal effect on total CC impact [68]. Across the reviewed literature, both microand macro-level studies confirm that incorporating geographically refined data can influence environmental impacts of wind energy. At the micro-level, studies of individual turbines or small-scale wind parks report CC impacts ranging widely, from as low as 4.5 gCO2eq/kWh in high wind-speed regions [62] to as high as 51.5 gCO2eq/kWh in desert areas with increased transport requirements [59]. These results highlight how site-specific conditions such as wind speed, transportation distances, foundation type, and maintenance logistics can influence the environmental impact. At the macro-level, regionalized LCAs of entire wind fleets or wind farms also show considerable variation. For instance, CC impacts of Chinese onshore wind were found to range between 19.88 and 28.2 gCO2eq/kWh [51,52]. Sacchi et al. (2018) assessed both onshore and offshore turbines in Denmark using a parameterized, spatially explicit LCA model and reported CC impacts ranging over time from 45.3 to 11.7 gCO2eq/kWh, depending on turbine size, location, and lifetime [66]. For offshore wind, additional 2.1. LCAS OF WIND ENERGY 19 studies such as Feng and Zhang (2023) reported CC impacts ranging from 13.3 to 29.5 gCO2eq/kWh, influenced by factors such as water depth, foundation type, and transportation distances [68]. Importantly, studies that apply geographically refined LCI data, incorporating parameters such as shore distance, water depth, air density, regional electricity mixes, and localized supply chains, tend to yield more accurate and context-specific results than those relying solely on normalizing impacts by local electricity yield. For example, Huang et al. (2022) and Sacchi et al. (2018) incorporate several geographically refined LCI data, such as water depth, shore distance, wind speed, air density, and transport distances [61,66]. Including these location-specific factors allows for more representative environmental assessments and enables better comparisons between turbine technologies and installation sites. These methods contrast with simpler regionalization approaches, which adjust only for variations in annual electricity production while assuming uniform upstream LCI data. In conclusion, integrating geographically explicit LCI data improves the representativeness and precision of LCAs for wind energy. While it introduces additional modeling complexity, this approach enables better informed decisions regarding turbine siting, supply chains, and environmental trade-offs, especially in the context of expanding wind fleets and diverse deployment conditions across regions. Table 2.1 presents a summary of the main characteristics of the included studies, whereas a full description together with the derived RQ is presented at the end of this Chapter in Section 2.5.1. 20 CHAPTER 2. LITERATURE ON EMERGING ENERGY TECHNOLOGIES Table 2.1: Overview of LCA studies on wind energy, highlighting the geographical scape, the turbine type, the foundation type and the regionalization parameter. The different scales refer to either single turbines (micro) or group of turbines, clustered in a fleet or farm (macro, Scope = Geographical scope of the study). Source Scope Turbine type Foundation type Scale Regionalization parameter [61] CN Offshore Monopile Micro Water depth, air density, wind speed, transmission & transportation [53] CN Onshore n.a. Macro Transport infrastructure [65] FR Offshore Semi-submersible floating Macro Transport, cables & electricity production [52] CN Onshore n.a. Macro n.a. [56] Global Offshore monopile, tripot, jacket Macro n.a. [57] PT Onshore n.a. Micro n.a. [45] CN Onshore n.a. Macro n.a. [54] CN Onshore n.a. Macro n.a. [67] DK On-& offshore Monopile Macro Wind speed, sea depth, shore distance [58] US Onshore n.a. Micro Transport [59] CN Onshore n.a. Micro Transport [66] DK On- & offshore Monopile Macro Cable length, market & time adjustments [55] CN Offshore Monopile Macro n.a. [64] US Offshore Monopile, tripod, floating Micro Wind speeds, water depths, grid connection [60] DE Offshore Jacket Micro Shore distance, water depth [63] EU Onshore n.a. Macro Electricity production [62] EU Onshore n.a. Macro Electricity production [68] CN On- & Offshore Monopile, jacket Macro Wind speed, distances [51] CN Onshore n.a. Macro regionalized capacity factors 2.2. LCAS OF EVS PROVIDING FLEXIBILITY SERVICES 21 2.2 LCAs of EVs providing flexibility services 2.2.1 Electric vehicles and their batteries Since the entry of EV in the mass market in the last decade, substantial research effort is dedicated to applying the LCA methodology to EVs in order to understand their environmental impacts [69]. Many studies aim to understand the environmental advantage of an EV over other drivetrain technologies [69– 71]. Over time, those studies enhanced their level of detail to increase their reliability and to represent reality as good as possible, for example by introducing a range-based assessment to include variability in LCA of EVs [72]. Additionally, the LCA studies of EVs are applied in different countries in order to cover the environmental impacts during the use stage [73]. At the same time, many LCA studies on the environmental impacts of the battery are published. The first and most well-known studies publishing primary LCI data are from Zackrisson et al. (2010), Majeau-Bettez et al. (2011) and Ellingsen et al. (2014) [74–76]. Until now, those studies are frequently used as a references for modeling mobile battery storage. Since those studies were published, more and more papers have been published evaluating different battery chemistries, impact categories, use stage assumptions, system boundaries, etc. A comprehensive overview of LCA studies on lithium-ion batteries (LIB) is provided by Peters et al. (2018). In their review paper, they analyze 79 different studies and summarize their results [77]. They highlight a great variance in GHG emissions, even though the LCI data originate mostly from the same studies, and found the average GHG emissions for the battery production of various chemistries are 110 kgCO2eq per kWh storage capacity. This variation can be attributed to differences in battery chemistries, which influence key performance characteristics such as energy density, cycle life, and thermal stability. Common chemistries include lithium-ion battery with lithium iron phosphate as cathode material (LFP), lithium-ion battery with nickel manganese cobalt oxide as cathode material (NMC), and lithiumion batteries with nickel cobalt aluminum oxide as cathode material (NCA), each with specific advantages and drawbacks. For instance, LFP batteries typically offer longer lifetimes and improved safety, while NMC and NCA provide higher energy densities, making them more suitable for space-constrained or performance-focused applications [77]. The choice of chemistry ultimately affects the environmental performance and suitability of a battery for different use cases, such as stationary storage, electric mobility, or second-life applications. More recently, research on LCA of EVs is dedicated to understanding the environmental impacts of future electricity mix and different EoL management strategies, giving the traction batteries a second life and investigating new materials in the battery [78–80]. All presented studies focus on EVs or their traction batteries, with the majority of the studies conducting an attributional LCA (ALCA) perspective. As suggested in the UNEP report ‘Global Guidance Principles for Life Cycle Assessment Databases’ and widely adopted in the field, two main LCA modeling approaches are distinguished: Attributional and Consequential [81]. ALCA quantifies the share of global environmental impacts attributable to a product’s life cycle [26, 81]. It assigns input and output data to the system’s function, representing the current state of the system [82]. In contrast, CLCA models the environmental impacts resulting from a decision-induced change in demand for a product or service [26,81,82]. 2.2.2 Assessing flexibility services in existing literature Quantifying environmental impacts of flexibility services, e.g. smart charging, poses several challenges: first, the proposed solution does not imply additional infrastructure, apart from additional power electronics at the charger or electric vehicle level, to allow bi-directional flow [83]. Thus, at first glance the environmental impacts of such services seem equal to EVs being charged uni-directional. However, when an EV is providing smart charging, e.g. powering a vehicle and feeding back electricity to a location, this is understood as a multi-functional process in LCA terminology. In LCA, processes or activities provid- 22 CHAPTER 2. LITERATURE ON EMERGING ENERGY TECHNOLOGIES ing two or more services or products are referred to as multi-functional processes. In order to understand the environmental impacts of such multi-functional processes, one way to address the challenge of multifunctional processes is the allocation of environmental impacts of the battery production of the EVs to the two provided services. According to ISO 14044, there are various ways to model multi-functional processes, such as system expansion, allocation and substitution [84,85]. Another argument to allocate environmental impacts to smart charging of the EV is the potential impact on the battery capacity. The additional charge and discharge behavior due to smart charging can pose additional stress on the battery and thus can reduce its lifetime. However, a lack of scientific confirmation leaves this research area open for further exploration [83]. Second, modeling the use stage of an EV providing any type of flexibility service requires, due to its complex interaction at the charging point, a more comprehensive modeling of the energy system. 2.2.3 Environmental impacts of flexibility services Table 2.2: Overview of LCA studies addressing flexibility services provided by electric vehicles (EVs), summarizing scope, time horizon, flexibility type, system level, and energy system model (ESM). Detailed descriptions of the LCA types conducted per study and other findings are provided in Appendix B in Table B.1. Source Scope Time Flexibility type Level ESM [86] EU 2050 Uni-& bidirectional Macro PERSEUS-EU model [87] DE 2050 V2G Macro Own [88] DE 2050 Demand-side mgmt. Macro Own [89] CN n.a. Uni-& bidirectional Macro Own [90] IT 2030 Uni-& bidirectional Macro EnergyPLAN [91] UK 2050 V2G Macro n.a. [92] DE n.a. Smart-& bidirectional Micro n.a. [93] GB n.a. Smart Micro n.a. [94] DE 2050 Uni-& bidirectional Macro-Micro ISaR First studies have begun to assess the environmental impacts associated with EVs that provide smart charging services. For example, Wohlschlager, Haas, and Neitz-Regett (2021) compare the impacts of information and communication technologies used in smart versus standard EV charging [92]. As the focus of the study is on comparing the infrastructure of the two different charging strategies, the functional unit is defined as enabling the charging of a private vehicle for one year, where the system boundaries include the intelligent metering systems as well as private wallboxes. The EV and its battery are excluded from the study. The study highlights that the environmental impacts of the infrastructure required for smart charging exceeds the impacts of uni-directional standard charging by 84 % [92]. Besides this study, Tang et al. (2021) compares the impact of smart and standard charging at British households [93]. Here, 2.2. LCAS OF EVS PROVIDING FLEXIBILITY SERVICES 23 the functional unit is set as single domestic EV charging event and a cradle-to-grave approach including the EV is selected. The authors found, that compared to standard charging, smart charging reduces CC impacts by 6 % based on most frequent charging behaviors, described by overnight charging [93]. Both of the studies are conducted for a household representing a micro-level assessment. Next to the impact of smart charging, modeling an energy system including EVs providing other flexibility services is until now done mostly from a macro-level perspective: Xu et al. (2020) for example study different EV charging strategies, including vehicle-to-grid (V2G), and implied environmental impacts of the EU energy system in 2050. They integrated ALCA into an ESM and found, that V2G services in 2050 could reduce GHG emissions by as much as 47 % [86]. The environmental impacts of the battery production used in the EVs to provide the V2G service remain unaddressed. Furthermore, the modeling considers static EV profiles to quantify the bi-directional charging. Another example is provided by Wang et al. (2021): they analyze different aggregation methods of EVs providing flexibility service into an ESM to understand the implications on the environmental impacts [87]. While they take into account electricity stored by the battery of the EVs, a specification on the allocation of the environmental impacts of the battery is absent. Furthermore, Nilges et al. (2024) present EVs as assets for demand-side management, defined as load-shifting of demand to hours with low electricity prices, in order to understand the consequences on the GHG emissions [88]. However, they do not use LCA to quantify the environmental impacts of the future German energy system integrating EVs. Another study investigates singleand bidirectional charging strategies of EVs and found the three following flexibility benefits: the bi-directional charging strategy is found to (i) reduce the total charging costs, (ii) improve the utilization rate of renewable energy shares and (iii) reduce carbon emissions [89]. In this study, obtained carbon emissions from another LCA study are monetized by a carbon price and together with the economic costs formulated as a minimization problem in the objective function of the ESM. How carbon emissions are exactly defined remains unclear. At the same time, the paper seems to overlook a clear difference between carbon and carbon dioxide emissions. Apart from the presented studies, which rather focus on the technical modeling of EVs and their flexibility potential from a macro-perspective, first attempts are made towards identifying environmental impacts of the flexibility services by applying LCA. In their study, Rovelli et al. (2021) coupled consequential LCA (CLCA) with an ESM called EnergyPLAN to understand consequences of EVs’ use stage [90]. Even though they limit their study to uni-directional charging, they claim that without additional installation of RET capacities, the future electricity demand of EVs might increase the CC impact by up to 40% compared to their business-as-usual (BAU) scenario. Similarly, Zhao et al. (2022) assess EVs as stationary storage in the United Kingdom in 2050 and the resulting effects on the environmental impacts [91]. In their study, they present four different future options for EV batteries considering different battery chemistries: a) flexibility services, b) battery swapping, c) reuse and d) stationary battery storage. For flexibility services, they found, compared to 0.165 kgCO2eq/kWh in 2018, the carbon footprint of the energy system can be reduced to 0.036 kgCO2eq/kWh in 2050. The study is conducted without making use of an ESM and thus, EVs and their flexibility potential is considered in a static manner, neglecting dynamic behavior of EV drivers. Contrary to existing studies evaluating the macro-level perspectives of EVs providing flexibility services, Wohlschlager et al. (2024) published a study that aims to reveal environmental effects of V2G charging in the future German energy systems next to evaluating impact on the battery of those EVs [94]. As starting point, they used the European linear optimization multi-ESM and coupled it with PLCA to obtain the hourly grid electricity mix in 2050. The PLCA focuses on revealing the difference in terms of CC impacts between uncoordinated and bi-directional charging of EVs. Additionally, they introduced two levels of assessment: a) the environmental impacts of the complete energy system is calculated and b) the environmental impacts on the technology level, e.g. on the EVs are determined. On a system level, V2G charging does not result in extremely high reduction of total GHG emissions by 2045. However, it speeds up the decarbonization of the electricity system between 2030 and 2035. On a technology level, V2G is 24 CHAPTER 2. LITERATURE ON EMERGING ENERGY TECHNOLOGIES found to reduce operational emissions in 2030 by between 50% up to 200% compared to uncontrolled charging. Due to the decarbonized energy systems, the V2G potential of operating GHG emissions will be limited, highlighting the importance of the production stage. Underlying methodological differences and scopes hinder a direct comparison of results of the presented studies, risking drawing wrong or misleading conclusions. For this reason, comparing results of the above presented literature is refrained within this thesis. Instead, the methodological differences of the analyzed studies are recapped in Table 2.2, supplemented with more LCA-relevant data such as type of LCA, functional unit, system boundaries, LCIA method and assessed impact categories. This Table B.1 is available in the Appendix B. Despite the methodological diversity and limited comparability across studies, some general conclusions can be drawn regarding the environmental implications of EVs providing flexibility services. Overall, the reviewed literature indicates that using EV batteries for grid flexibility during their primary life, through strategies such as unidirectional smart charging or bidirectional V2G, has the potential to reduce environmental impacts, particularly CC impacts, under specific conditions. These benefits are most pronounced when flexibility services align with periods of high renewable electricity generation and when additional renewable capacity is deployed to meet increasing electricity demand. Among the reviewed strategies, bidirectional charging is consistently shown to offer greater environmental benefits than unidirectional or uncontrolled charging. Wohlschlager et al. (2024) report that V2G can reduce operational CC impacts by up to 200% compared to uncontrolled charging, due to its ability to shift electricity use to hours with low CC impacts and displace the need for stationary storage[94]. However, as electricity systems decarbonize, the relative contribution of V2G to environmental impact reduction decreases, and the production phase of EVs and batteries becomes more prominent in total life cycle impacts. Unidirectional smart charging also shows environmental advantages compared to uncontrolled charging, primarily by enabling better alignment of charging demand with renewable electricity availability. However, its mitigation potential is generally lower than that of bidirectional charging, as it does not enable discharging back to the grid. In summary, flexibility services provided by EVs during their first life are generally found to reduce environmental impacts when implemented with appropriate system-level considerations. Among these, V2G shows the highest potential for CC impact mitigation, particularly in medium-term energy transitions. A concise description of the available literature on LCA of flexibility services provided by EVs is provided in the Section 2.5.1 along with the resulting RQ. 2.5. SUMMARY OF EXISTING LITERATURE AND RESEARCH GAPS 31 2.5.3 Claim for new models Section 2.4 presents that LCA has been widely employed to evaluate the environmental performance of PV installations, battery systems, and integrated energy networks across various scales and geographies. Despite a growing amount of literature, most studies remain limited in scope, either technologically (e.g., assessing only PV installations or only batteries), methodologically (e.g., excluding energy modeling), or in terms of assessed impact categories (e.g., focusing on CC impacts only). Studies integrating energy system models into LCA are relatively few, although such integration offers a more accurate representation of energy flows, operational dynamics, and system interactions. Likewise, prospective studies assessing future scenarios or technology life cycles remain rare. While some efforts have emerged to assess system configurations in Belgium and beyond, a comprehensive, forward-looking environmental assessment of decentralized energy systems remains unaddressed. Finally, the Research Gap 4 followed by identified Challenges can be posed as follows: Research Gap 4: No existing demonstration of the evolution of environmental impacts from a national roll-out of renewable energy systems supported by stationary storage considering different flexibility criteria in Belgium. 4.1 Integrating ESM & LCA for BE 4.2 Combining top-down macro assessments with bottom-up micro LCAs 4.3 Evaluating future system dynamics in a holistic and forward-looking manner First, many LCA studies assess PV installations or battery systems in isolation, failing to capture the combined environmental performance of integrated energy systems that include both generation and storage components. Second, energy system modeling is often not incorporated into LCA studies, despite its potential to provide more realistic input data and system interactions. Third, most studies focus exclusively on the CC impacts, underrepresenting other relevant impact categories. Fourth, although some studies acknowledge long system lifetimes, very few adopt a prospective approach that reflects evolving energy contexts or future grid mixes. Lastly, within the Belgian context, few studies provide a geographically and technologically comprehensive LCA of decentralized energy systems. This study addresses these gaps by conducting a prospective, integrated LCA of decentralized PV-battery systems, accounting for spatial, temporal, and technological dimensions relevant to the Belgian energy transition. Section 1.4, 1.5, 1.6, and 1.7 describes, how this thesis builds on the identified gaps. 32 CHAPTER 2. LITERATURE ON EMERGING ENERGY TECHNOLOGIES Chapter 3 Methodology for assessing emerging energy technologies 3.1 Overall methodology Current life cycle assessment Following DIN ISO 14040, LCA consists of four stages: 1.) goal and scope definition, 2.) LCI analysis, 3.) LCIA and 4.) interpretation [24]. The goal and scope phase defines the system boundaries, functional unit, and impact assessment method. The LCI phase involves collecting data on inputs from and outputs to the environment, covering both foreground and background systems. In the LCIA stage, the collected inventories are translated into impact results for selected environmental categories, followed by the interpretation of those results. Figure 3.1 outlines the structure of the method section, detailing goal and scope, LCI and LCIA phases of this thesis. This analysis is structured across four main data axes: refined geographical inventories (Section 3.2), refined temporal inventories (Section 3.3), inventories for small and medium scale (Section 3.4), and inventories for large scales (Section 3.5). Each outlined Section begins with the definition of goal and scope, along with the respective description of the different data axes. Chapter 3 ends with a description of applied LCIA methods and other compiled metrics in Section 3.6, whereas Section 3.7 details the evaluation of uncertainty and parameter sensitivities. In general, two main types of LCA can be distinguished: ALCA and CLCA. These differ primarily in the questions they address, how they treat multifunctionality, and how underlying data are modeled. ALCA, often described as an “accounting” or “descriptive” approach, aims to identify which share of global environmental impacts can be attributed to a specific product. The system model includes processes connected through physical flows (e.g., energy, materials, waste) that supply the functional unit [130]. The typical question addressed by ALCA is “What environmental impacts can the studied product be held responsible for?” [131]. Multifunctionality in ALCA is typically resolved through allocation or, less frequently, system expansion [132]. Moreover, data in ALCA are modeled using average process or market data [131,132]. CLCA, in contrast, represents a change-oriented approach that investigates environmental consequences directly or indirectly caused by a decision. A CLCA considers only those processes that are affected by a change in demand or supply, thereby modeling the cause-and-effect chain resulting from a decision [130]. The key question addressed by CLCA is “What are the environmental consequences of consuming one additional unit of a product?” [132]. Multifunctionality in CLCA is handled using 33 34 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES Figure 3.1: Stages of an LCA, emphasizing the refined LCI. The numbers in parenthesis represent the Section numbers where those topics are explained in detail (following DIN ISO 14040 [24]). system expansion, and data are modeled using marginal information reflecting technologies affected by an incremental change [131,132]. Based on these definitions, different types of LCA could be justified for the various assessments presented in this thesis. When evaluating individual technologies, such as a wind turbine or SLB of specific size, an attributional approach is appropriate to determine the environmental impacts associated with producing or operating that technology. Conversely, when introducing decentralized PV installations combined with flexibility solutions, the question of what environmental consequences result from this decision could be framed in a consequential context. A general misconception, however, is that studies at larger scales necessarily require CLCA, as their results inform decision-making. Schaubroeck et al. (2021) [133] emphasize that this association is not inherent to the scale of the analysis but to the type of question addressed. Furthermore, they highlight the lack of clear guidance on how to consistently distinguish between ALCA and CLCA in practice. In this thesis, the central research question concerns how much environmental impact can be attributed to specific technologies or services under varying future system conditions. According to the formal definitions, this aligns with an attributional perspective. A decisive methodological consideration supporting the choice of ALCA is the treatment of multifunctionality. Throughout the thesis, multifunctional processes are handled by allocation, while system expansion is deliberately avoided. Allocation is suitable 3.1. OVERALL METHODOLOGY 35 given the detailed modeling of assets and the exploration of several allocation scenarios to reflect possible variations in co-product relationships. According to established guidelines, allocation is consistent with ALCA, whereas system expansion is typically applied in CLCA [132]. This supports the methodological coherence of the attributional approach applied here. Another relevant criterion is the representation of the background system. Following Schaubroeck et al. (2021) [133], the conventional dichotomy between “average” data for ALCA and “marginal” data for CLCA is not always clear-cut. Their analysis reveals multiple interpretations of “marginal” and questions the strict separation between average and marginal data use. Consequently, employing future-oriented average supply mixes, based on projected capacity developments and technology shares, can still be consistent with an attributional framework. CLCA typically quantifies the environmental consequences of a change in demand. For electricity systems, this involves identifying the marginal supply mix that meets one additional kilowatt-hour of consumption. In an expanding electricity market, marginal suppliers are the most cost-competitive technologies, whereas constrained technologies are excluded from the marginal mix, as those are unable to supply an additional unit of the product. Following the CLCA logic, this means that additional electricity demand is entirely met by wind power. However, this interpretation of “marginal” differs from its use in energy system modeling, where the term refers to hourly marginal mixes based on the merit order. Depending on the selected hour, the marginal supplier may vary (e.g., gas turbines at peak hours). Thus, applying a strict CLCA approach to a dynamic electricity system remains challenging. In this thesis, the focus lies on identifying attributable environmental impacts of technologies within a future-oriented context. Average market mixes are modeled according to the projected generation shares and their resulting adjusted merit order for future years. Although this approach integrates elements typically associated with consequential reasoning, such as future capacity evolution based on a given policy context, it remains attributional in nature, as the aim is to assign environmental responsibility, not to model system-wide consequences of decisions. Hence, this work can be understood as a futureoriented ALCA, acknowledging that the consideration of future changes in generation capacities and resulting supply mixes may reflect a CLCA-inspired perspective within an ALCA framework. Finally, it is worth noting that the majority of studies integrating ESM with LCA have adopted an attributional approach to assess electricity generation [134]. This further supports the methodological alignment of the present work. Additional reflections on this choice are provided in Section 7.4. Future life cycle assessment The technologies analyzed in this thesis vary in terms of technological maturity. Wind turbines have been commercially available for decades, and continuous innovation has enhanced their performance. In contrast, discharging EVs as a grid service is a relatively novel but market-ready concept, although not yet widely implemented [83]. While the service maturity of EV discharging is high, the product maturity of SLB remains low. SLB used for stationary storage are currently primarily limited to demonstration projects. Their full potential and commercial deployment timeline remain uncertain. In any case, SLB will not enter stationary applications immediately, as their availability is delayed by the first-use phase in EVs, which in Belgium typically lasts around nine years [23]. Given this time lag, a PLCA is conducted to assess SLB, aiming to model their environmental impacts at a future point in time, relative to the study’s commissioning date [25]. PLCA enables an understanding of how CC impacts may evolve due to technological advancements in both the foreground (e.g., production improvements) and background systems (e.g., decarbonization of electricity supply) [135]. For this purpose, the Python library ‘premise‘ (version 1.8.1) [42] is used to transform the ecoinvent 3.9.1 cut-off database [46] based on scenarios from the integrated assessment model REMIND [136]. In addition to SLB, PLCA is also applied to assess the future environmental performance of currently commissioned wind energy technologies. The Section 3.6 details why the focus in PLCA lies on CC 36 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES impacts and no other impact categories are considered. Further information on the selected IAM and the rationale for choosing the REMIND model is provided in Section 3.3. 3.2 Refined geographical inventories 3.2.1 Goal and scope definition The objective of this LCA is to develop a model to assess onand offshore wind turbines all over Europe, taking into account the turbines’ specific installation location. The functional unit is set to be 1 kWh delivered electricity. To obtain geographically refined LCI, an existing open-source model developed to assess the Danish wind turbine fleet producing cradle-to-grave inventory using dynamic scaling functions is used as starting point and further developed to assess wind turbines on the European continent [66]. Although the model is capable of assessing wind turbines all over Europe, the geographical scope within this thesis is limited to Belgium only. While the updated model itself is available on Github, all generated LCI data are processed directly in the LCA model itself [43]. Consequently, no additional LCI data is generated. Figure 3.2 depicts the system boundaries of the wind power assessment (PAPER 1). Figure 3.2: System boundaries, refined LCI data integration and general modeling flow chart to assess wind turbines with geographically refined life cycle inventories (lon = Longitude; lat = Latitude; ISO = International Organization for Standardization; LCA = Life cycle assessment; LCI = Life cycle inventory). Apart from updating the latest Brightway version with improved and additional functionalities and changing the background database to the ecoinvent version 3.9.1. cut-off [46, 137], the following five location-specific parameters are improved: 1. Type of installation, 3.2. REFINED GEOGRAPHICAL INVENTORIES 37 2. Foundation type, 3. Distance to the electric power grid, 4. Transport distances and 5. Electricity production of the turbines Results are presented in two steps: first, the impacts on a single-turbine level are presented, followed by an evaluation of the Belgian wind turbine fleet. More detailed explanation on the LCI data of the turbine is available in the public WIMBY deliverable ‘D2.8 LCA results report and spatially explicit data (b)’ [35]. 3.2.2 Geographically refined inventories Activities related to wind turbines do not all happen at the manufacturing location of the wind turbines. During the assembly, maintenance and disposal stages, the required materials and activities can be sourced locally, e.g. electricity in the assembly stage. The electricity consumed during the assembly of the wind turbine at its future location is assumed to be sourced from the country where the turbine is installed. Such dynamics are implemented in the LCA model in a more variable way, so that the location of the datasets in the assembly, maintenance and disposal stage changes if the country of the wind turbine is modified. Contrary to local sourcing, the material sourcing in the component production stage is unlikely to be supplied locally, but rather originate from a global or European market. At the other end of the life cycle, the EoL treatment is adopted unchanged from Sacchi et al. (2019), treatment options are included based on the type of materials [66]. Concrete, fiberglass and aggregates end up in landfills, while steel, thermoplastics and copper cables enter scrap markets [66]. The problematic of treatment options of wind turbines, especially of the wind turbine blades is further outlined in the first WIMBY deliverable ‘D2.7 LCA results report and spatially explicit data (a)’ [34]. Foundation types As the original “WIND_LCA_DK” model already includes monopile foundation, which can be installed in water depths of up to 30 meters [64], two more offshore foundations are added to the WIMBY LCA model. Thus, the model chooses different foundation types based on the sea depths. The sea depths is chosen as the only decision criterion, even though this might not fully represent reality. However, as other decision criteria, e.g. types of seabed are more difficult to convert into a parameter that can be included in a simple manner in the LCA function, currently the sea depth serves as only parameter for the selection of offshore foundations. To use the sea depth as decision criteria for the floating foundations, the sea depth is first obtained using the turbines location and GEBCO’s global gridded bathymetric data sets [138]. Depending on the sea depth, the following offshore foundations are selected: • 0-30 meters sea depth: monopile • 30-60 meters sea depth: semi-submersible floating foundation • Deeper than 60 meters: spar buoy floating foundation Due to size and location similarities, LCI data for semi-submersible floating foundations (semi-sub) and spar buoy floating foundations (spar) are adopted from Garcia-Teruel et al. (2022) [139]. For the spar, the reported rated power is 9.5 MW, whereas the turbine with a semi-sub has a rated power of 6 MW. In terms of functionality of the semi-sub, the turbine is mounted on a floating platform, which is attached to the seabed via 4 anchors at a depth of between 60-80 meters. The spar concept is an extension of the 38 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES turbine tower, similar to a tension leg, filled with a ballast (in this case: iron ore). The spar buoy is fixed to the seabed via three anchors. This floating solution is reported to be in place at sea depths between 95 and 120 meters. Figure 3.3 depicts the three foundation types modeled in the scope of this thesis. Figure 3.3: Offshore wind turbine foundation types depending on the sea depth (Adapted from Yordan et al. (2024) [140]). Cable lengths Datasets of a topologically connected presentation of the European high-voltage grid based on OpenStreetMap are consulted and used to determine the cable length required to connect the wind turbines to the grid [141]. In this public repository, datasets for buses, converters, transformers, etc. are available. In the absence of substations, the electric power buses datasets are used as proxy and for further processing. Electric power buses are nodes connecting several lines including several components, e.g., loads and generators. With Pandas ‘geopy’ package, the geodesic distance between the turbine location and the closest power bus is calculated and used to determine the cable length in the LCA model. Transport distances Sub-component production locations of the two largest European wind turbine manufacturers, VESTAS and ENERCON, together accounting for 55 % of all installed turbines, are investigated [35]. A function is defined to identify the shortest, accumulated distance of all the components from the turbine location. If a component can be manufactured at more than one facility, the one closest to the wind turbine location is chosen. Distance for rotor, nacelle and tower are kept variable, while for the foundation it is kept constant (50 km). The distances are then included in the LCA model for transportation impacts. Electricity production The electricity production used in this assessment is obtained in three different ways: for the case studies, the electricity production is obtained from the WIMBY map directly. The WIMBY map is the product of the WIMBY project, which is an interactive map to place turbines all over Europe and obtain their techno-environmental and social impacts. No own calculations of annual electricity production for the case studies are done. The rated power, AEP and CF of the Belgian onand offshore turbine are 3.4 and 3.2. REFINED GEOGRAPHICAL INVENTORIES 39 10 MW, 9.7 and 51.6 GWh and 33 and 56 % respectively. Those values are directly extracted from the WIMBY interactive map. The locations are randomly chosen on the WIMBY map and do not represent real installations. Instead, these case studies are intended to depict potential wind turbines. For the evaluation of the current Belgian wind fleet, the electricity production is calculated independently. It is important to highlight, that the calculations are just an approximation and do not represent real electricity production of the Belgian wind turbine fleet. In short, the calculation can be described as follows: The annual electricity production of each Belgian wind turbine is calculated using an updated turbine register, average wind speed data with bias correction from the Global Wind Atlas, and a parametric power curve model by Saint-Drenan et al. (2020), based on rated power and rotor diameter [142]. Standard cut-in (3 m/s) and cut-off (25 m/s) wind speeds are assumed, which are matched to turbine hub heights and locations, rounded to the nearest ERA5 grid point. Hourly wind speed data are used to calculate daily power output, which is scaled to monthly and annual values [143,144]. Within the WIMBY project, an internal register of all European wind turbines is provided, which is not available due to licensing issues. CFs are derived to validate results against EU onshore and offshore fleet averages for 2022 (24 % and 34 %, respectively [145]). Based on the existing data, the future Belgian wind fleet evaluation focuses on a more theoretical approach, considering projected rated power and calculated, projected future CFs. 40 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES 3.3 Refined temporal inventories 3.3.1 Goal and scope definition The objective of this assessment is to quantify the environmental impact of retired EV batteries that are repurposed for a second-life in Belgium. Consequently, the FU is set to one kWh delivered electricity by the battery. A cradle-to-grave approach is selected for the system boundaries to primarily evaluate the CC impact for SLB in Belgium, excluding the first life (Figure 3.4). Further details about applied LCIA categories other than CC impacts are provided in Section 3.6. For comparison reasons, benchmark batteries based on the same battery chemistry as the SLB, manufactured specifically for stationary electricity storage in the respective use cases and directly recycled at their EoL are included. This approach is chosen to allow a comparison of SLB and the benchmark, excluding implications of different technical system designs. Figure 3.4: System boundaries for the second-life and the benchmark battery including the time dimension (RES= Residential use case, IND= Industrial use case, UTI= Utility use case). System boundaries Figure 3.4 presents the design of the study with a temporal dimension, the first and second use stages and the different use cases for the stationary storage for the SLB and the benchmark batteries. The batteries used as traction batteries in their first life are produced in 2023. According to Belgian national statistics on the average age of vehicles in 2022, the batteries reached their first EoL nine years later [23]. In 2032, the retired batteries will be removed from the EVs, collected and tested. For this study, EV batteries that are not suitable for second use are considered to be out of the scope. If deemed suitable for second use, the batteries are dismantled, repurposed and, depending on their subsequent use case, additional components, e.g. casings or a new BMS will be added. Consequently, the SLB will function as stationary storage in three different use cases: 1) residential, 2) industrial and 3) utility use cases. After their second use stage, the SLB will be once more collected and, finally, recycled. Accordingly, the described system boundaries include stages linked to both the first and the second life: repurposing, stationary 3.4. INVENTORIES FOR SMALL AND MEDIUM SCALES 47 At the micro-level use case, a physical allocation based on delivered electricity for stationary storage and driving over the battery’s lifetime is determined. In the absence of long-term studies investigating the performance over both first and second-life of EV batteries, no degradation is taken into account in the current study. This allocation factor allows accounting for the environmental burden of manufacturing the battery in an EV when modeling systems where smart charging of EVs is performed. Data of electricity charged and discharged are obtained from the framework described in Section 3.4.2. The allocation factor is computed with Equation 3.2: AFflex =Estat (Estat +Emobile)(3.2) Where: • AFflex: allocation factor for providing flexibility services (%), • Estat: discharged electricity for providing flexibility services from the EV battery (kWh/year), • Emobile: discharged electricity for powering the EV (kWh/year). The allocation of EV battery impacts at the mezzo-level use case is more complex: Every charging session in theory can be initiated with another EV. Thus, summing up the discharged and charged electricity per charger is not feasible in the mezzo-level use case. Instead, the battery manufacturing impacts are allocated to a separate EV for every discharging event. That means that every time electricity is discharged, the manufacturing impacts of another EV with different chemistry and EV battery size is considered. The total electricity charged is calculated for every EV individually based on assumptions and national statistics. In the simulation, each recorded charging session represents a unique EV. Unfortunately, the battery chemistries of the full Belgian EV fleet are not publicly available. As a proxy, the shares of the EV battery sales of 2024 are utilized. Battery chemistries are assigned stochastically based on 2024 EV sales market shares (see Table 3.4), and battery capacities range from 40 to 100 kWh [153]. Table 3.4: Battery chemistries and their market shares based on 2024 EV sales in Belgium (LFP = Lithium-ion battery with lithium iron phosphate as cathode material; NMC111 = lithium-ion battery with lithium nickel manganese cobalt oxide as cathode material (LiNi0.33Mn0.33Co0.33O2); NMC811 = Lithium-ion battery with lithium nickel manganese cobalt oxide as cathode material (LiNi0.8Mn0.1Co0.1O2); NCA = Nickel cobalt aluminium oxide; source: [153]). Battery Chemistry BE Market Share in 2024 (%) LFP 11 NMC111 13 NMC811 38 NCA 38 Each chemistry–capacity pair is linked to a specific manufacturing impact (kgCO2eq). Energy consumption per kilometer is capacity-dependent (own approximations: 0.15–0.18 kWh/km), and a total lifetime milage of 139,587 km is assumed [23]. These inputs allow calculation of a lifetime energy throughput per EV, over which the battery impact is allocated, yielding a normalized value in kgCO2eq/kWh. This value is then multiplied by hourly discharged electricity to attribute impacts dynamically. 48 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES 3.4.2 LCI data collection of small and medium scales Design and optimization framework To model the four different system configurations at the microand mezzo-level use cases, the size and optimization are simulated using a VUB internally developed design and optimization framework. The framework aims to simulate and optimize the design and operation of an energy system, based on different electric assets and quantifying its technical, economic and environmental performance. The core of the virtual framework builds an optimization algorithm to reproduce assets modeled based on state-of-the-art from literature and specifically developed models. First, the framework identifies the optimal size of each asset by investigating the power dispatch during the given time. As part of this optimization, various key performance indicators can be selected. In this study, the minimization of total cost, including investment and operation costs, is selected as objective function. A comprehensive explanation is published by Felice et al. (2022, 2024) [152, 154]. Additionally, PAPER 3 contains further information on the individual methodologies, but also their integration. While more detailed explanations on the framework within the scope of this work are not provided, the following Section 3.4.2 and 3.4.2 describe the input data to simulate the microand mezzo-level use cases with the framework. Micro-Level Assessment The micro-level use case describes a residential four-person household in Belgium. Example household consumption data on a 15 minutes time resolution for 2022 are obtained from smart meters of a randomly extracted profile in Flanders, provided by Fluvius, the Flemish electricity and gas grid operator [155]. Profiles without heat pumps, EV charging and PV panels are selected to feed the profile as input data into the DOF. The electricity consumption of the selected household is 3,898 kWh per year [155], representing well the average household consumption of 3,500 kWh/year in Belgium [156]. Additionally, it is assumed that one EV is charged at the micro-level use case with a battery capacity of 80.3 kWh with an 11 kW charger assuming a 95 % charging efficiency. This charging profile is extracted from Sørensen et al. (2021) [157]. Interpretation and visualization of the demand and supply of each scale can be found in Chapter 5. The electricity supply in the micro-level use case is provisioned by PV installations, consumption from the national grid or, in the case that BESS is available or smart charging is performed, electricity discharged from the BESS or EV battery. The PV installation is simulated over a lifetime of 25 years, available at a capital expenditure (CAPEX) of 1,000.00 EUR/kWp and has operating costs of 7.50 EUR/kWh. The price of one kWh grid electricity is 0.40 EUR/kWh [158] and has an annual average CC impact of 152.49 gCO2eq/kWh in 2023 (source: own calculations). In the case of BESS installations, the price per kWh capacity is 433.00 EUR and a charge/discharge efficiency of 95 % is used [159]. In the micro-level use case, only one BESS based on an LFP chemistry is assumed. Mezzo-Level Assessment The mezzo-level use case represents a larger area, comparable to an industrial or commercial site. Data are obtained as described in PAPER 3 and sum up to an annual consumption of about 3.4 GWh. PAPER 3also contains more detailed analysis of supply and demand of the Green Energy Park. These data do not represent all entities located at the Green Energy Park, but are only a selection of available data, which cannot be shared due to confidentiality. Instead of a single EV, the simulation assumes 30 charging stations, which are accessible to the public and not restricted to certain EV drivers. While some input data specified in the micro-level use case remained equal in the mezzo-level use case, others are modified and presented hereafter. A lower CAPEX for the PV installation of the mezzo-level use case is used, in particular 850 EUR/kWp [9]. Furthermore, in the mezzo-level use case a lower electricity price of 3.4. INVENTORIES FOR SMALL AND MEDIUM SCALES 49 0.30 EUR/kWh based on average prices for commercial customers in Flanders [158]. Additionally, a wind turbine is foreseen in the mezzo-level use case with a rated power of 2 MW, a hub height of 78 meters and a rotor diameter of 80 meters. Its CAPEX is set to 2.2 million EUR, with 5 % operating costs of the CAPEX. Its operation is expected to produce electricity over 25 years. Next to the CAPEX of the BESS, a maximum capacity limit of 400 kWh is set for both stationary batteries at the Green Energy Park. Variation of the input data on the DOF output is not part of this work and has already been demonstrated previously [154], and is therefore considered out of scope. Instead, the variation in charging strategies in combination with BESS results in different asset capacities and hence, different electricity generation and supply. The most important input parameters are summarized in Table 3.5. Table 3.5: Common parameters used in the LCA and DOF to simulate the microand mezzo-level use cases across four system configurations (BESS = Battery electric stationary storage; PV = Photovoltaic; WT = Wind turbine; EV = Electric vehicle; LFP = Lithium iron phosphate; NMC = Lithium nickel manganese cobalt oxide). Parameter Unit Uni – no BESS Uni – BESS Smart – no BESS Smart – BESS PV lifetime years 25 25 25 25 WT lifetime years 25 25 25 25 EV battery capacity kWh 80.30 80.30 80.30 80.30 EV battery weight kg 518.00 518.00 518.00 518.00 Energy density – LFP kWh/kg 0.12 0.12 0.12 0.12 Energy density – NMC kWh/kg 0.14 0.14 0.14 0.14 EV lifetime years 9.75 9.75 9.75 9.75 Project lifetime years 25 25 25 25 Lifetime LFP [16] years 19 19 19 19 Lifetime NMC [16] years 18 18 18 18 Other adjusted LCI data Due to a scope with a limited amount of assets, instead of using LCI data available in commercial databases, the LCI of this study for the microand macro-level use cases are modeled more precisely. These assets are: •The Belgian electricity grid mix: Therefore, data for the Belgian electricity grid mix, the PV installations and BESS are collected from literature. Reasons for choosing other LCI data instead of data from well-recognized databases such as for example ecoinvent depend on the technology: for the Belgian grid electricity mix, only an average value is available in ecoinvent. Using the CC method of the Environmental Footprint version 3.1 (EF v3.1) results in an average of 206 gCO2eq/kWh, whereas CC impact from the electricity map varies between 36 and 372 gCO2eq/kWh, making up an average of 170 gCO2eq/kWh in 2023 [160]. Moreover, as the objective of this study is to understand trade-offs per hour over one year, a higher time resolution than annual average values is needed. For this reason, the ’dynamic LCA model’ proposed by Naumann et al. (2024) is adjusted for Belgium [161]. In this approach, the authors calculate hourly CC impact of the German electricity grid mix, taking into account imports and exports from neighbouring countries. Additionally, their approach is normalized over five years to represent a generic year neglecting unique events such as the COVID time or the natural gas crises. The underlying data are also obtained from the Transparency Platform of the ENTSO-E. Three datasets 50 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES per country (Belgium and its importing/exporting countries) are obtained: the hourly-resolved actual generation per production type, the installed capacity per production type in annual resolution and the cross-border physical flows in hourly resolution [14]. As a result, the hourly CC impact per hourly timestep for the Belgium electricity consumption mix is calculated. Due to complex computation and as currently most popular environmental impact used for steering current policy, CC impacts are assessed as first indicator. The hourly environmental impacts of other categories are currently still under development. Until then, to calculate other impacts than CC impacts, the average electricity mix from ecoinvent is used. •PV installations: Instead of using existing datasets from ecoinvent for PV installations, the LCI data for PV installations are extracted from a report of the International Energy Agency under the title ’Methodology Guidelines on Life Cycle Assessment of Photovoltaic’ [162]. Supported by a yearly PV energy production of 2,996.39 kWh for a 3 kWp PV installation in Brussels, which is obtained from Photovoltaic Geographical Information System (PVGIS), the CC impact per kilowatt hour of produced electricity is determined [163]. The lifetime of the Si-based PV installations is estimated to be 30 years [164]. Outdated data, e.g. efficiencies dating back to 2005 in the PV generated electricity datasets of the ecoinvent database or updated production locations are one main reason for remodeling the installations [46,165]. Multi-Silicon PV installations with a capacity of 3 kWp are included, as this technology represents the dominating PV technology for small scale installations according to ecoinvent [46]. •BESS Currently, only lead acid stationary batteries are available in ecoinvent, whereas all lithiumion battery datasets are for mobile applications [46]. According to the International Energy Agency, the dominant technology for BESS in 2022 remained lithium-ion battery. Due to the absence of lithium-ion BESS in ecoinvent, LCI data are used from Le Varlet et al. (2020) [16]. Furthermore, Le Varlet et al. (2020) identifies LFP as one of the most represented BESS technologies. Therefore, the mobile battery LCI modified by Le Varlet et al. (2020) of LFP batteries are used in the microlevel use case. In the mezzo-level use case, one LFP and one NMC battery are selected as BESS, as those two chemistries are already installed at the Green Energy Park. 3.4. INVENTORIES FOR SMALL AND MEDIUM SCALES 51 3.4.3 Calculations Micro-level use case To quantify whether and how much environmental impacts the micro-level system has considering different charging strategies, the CC impact at the micro-level is calculated considering PV installation and electricity supplied by the national Belgian electricity grid mix, eventually supported by BESS. First, the total impacts are calculated, which are in Equation 3.8 then expressed in the FU. The CC impact of the national Belgian electricity grid mix is calculated applying Equation 3.3: CCMicro Grid = (∑ t CCGrid(h)×∑ t DMicro Grid (h))×LTTotal (3.3) Where: • CCMicro Grid : are the total CC impacts of the national Belgian electricity grid mix over system lifetime (gCO2eq), • CCGrid(h): are the CC impacts of the Belgian grid at a given hour (h)(gCO2eq/kWh), • DMicro Grid (h): is the consumed grid electricity at hour (h)(kWh), • LTTotal: is the project lifetime assumed in this LCA study in years. The CC impacts of the PV installations can be calculated following Equation 3.4: CCiMicro PV =SiMicro PV ×CCPV ×LTTotal LTPV (3.4) Where: • CCiMicro PV : are the CC impacts of the PV installation at the micro-level use case for charging strategy i(gCO2eq), • SiMicro PV : is the size of the PV installation at the micro-level use case when providing charging strategy i(kWp), • CCPV: are the CC impacts of the PV installations (gCO2eq/kWh), • LTPV: is the lifetime of the PV installation (years). For some charging strategies, the installation of a BESS is foreseen, in particular for standard unidirectional and smart charging. The CC impact of manufacturing the BESS can be determined applying Equation 3.5: CCiMicro BESS =SiMicro BESS ×CCBESS ×LTTotal LTBESS (3.5) Where: • CCiMicro BESS: are the CC impacts of manufacturing a BESS for charging strategy i(gCO2eq), • SiMicro BESS: is the size of the BESS at the micro-level use case when providing charging strategy i(kWh capacity), • CCBESS: are the CC impacts of manufacturing a BESS (gCO2eq/kWh capacity), • LTBESS: is the lifetime of the BESS (years). 52 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES Given the allocation factor defined in Equation 3.2, the CC impacts of the EV battery can be determined using the allocation factor as described in Equation 3.6: CCiMicro EVB =SiMicro EVB ×CCEVB ×AFflex ×LTTotal LTEVB (3.6) Where: • CCiMicro EVB : are the allocated CC impacts of battery installed in an EV in the micro-level use case (gCO2eq), • SiMicro EVB : is the size of the battery installed in an EV (kWh capacity), • CCEVB: are the CC impacts of battery installed in an EV (gCO2eq/kWh capacity), • LTEVB: is the lifetime of the battery installed in an EV (years). Next, Equations 3.3 to 3.6 are summarized to obtain the total CC impacts of the micro-level when considering various charging strategies in Equation 3.7: CCiMicro =CCMicro Grid +CCiMicro PV +CCiMicro BESS +CCiMicro EVB (3.7) Where: • CCiMicro: are the CC impacts of the micro-level use case considering different charging strategies i (gCO2eq). To facilitate comparison, the CC impact is then converted to its functional unit via Equation 3.8: UiMicro =CCiMicro EMicro ×LTTotal (3.8) Where: • UiMicro: are the CC impacts of the micro-level use case when providing different charging strategies i(gCO2eq/kWh), • EMicro is the generated electricity of the micro-level use case per year (kWh/year). Mezzo-level use case The mezzo-level use case differs from the micro-level use case by its scale: due to a higher electricity consumption, the assets increase in size, more electricity supplied by the national Belgian electricity grid mix is consumed and various EVs are considered at this scale. The same Equations 3.3-3.7 are applied to identify the environmental impacts at the mezzo-level, allowing conclusions related to upscaling of the investigated scale. Additionally, in the mezzo-level use case wind turbines are included, whose CC impacts can be calculated following Equation 3.9: CCiMezzo WT =SiMezzo WT ×CCWT ×LTTotal LTWT (3.9) Where: • CCiMezzo WT : are the CC impacts of the wind turbine at the mezzo-level use case for charging strategy i(gCO2eq), 3.4. INVENTORIES FOR SMALL AND MEDIUM SCALES 53 • SiMezzo WT : is the size of the wind turbine at the mezzo-level use case when providing charging strategy i(kW), • CCWT: are the CC impacts of the wind turbine (gCO2eq/kWh), • LTWT: is the lifetime of the wind turbine (years). Again, CC impacts of the mezzo-level are expressed in the functional unit of kWh generated electricity using Equation 3.7 adding Equation 3.9. 54 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES 3.5 Inventories for large scales 3.5.1 Goal and Scope Definition This Section presents a framework for assessing the environmental impacts of the Belgian energy system at a macro scale. Specifically, it aims to evaluate the impacts of decentralized technologies, including various levels of flexibility, on electricity supply in Belgium for the years 2030 and 2050. The FU is defined as one kWh of electricity consumed in the residential and commercial sectors for the respective years (i.e., 2014, 2030, and 2050), as supplied by the system under assessment. Results are published in PAPER 5 and PAPER 6. Even though in this context the premise tool is not applied, the study is still carried out in a prospective, future-oriented manner. Figure 3.6 illustrates the research design adopted for this assessment. The scenarios and associated energy balances are derived from a large-scale energy impact model and enriched with data from the literature on future electricity projections, refined LCI data, and data from commercially available databases. Based on the collected data, an LCA is performed for each reference year in accordance with the methodological approach detailed in Section 3.1. Figure 3.6: Research design integrating a TIMES-based energy model into the life cycle assessment framework. Thin arrows represent data flow; light grey arrows indicate iterative processes between the different stages of LCA (BE = Belgium; LCI = Life Cycle Inventories; EoL = End-of-Life; PV = Photovoltaic). System Boundaries This study evaluates the national implementation of specific decentralized technologies, such as PV installations and energy storage systems. Consequently, the analysis focuses on the main components, namely PV installations and storage systems, including inverters, while omitting smaller elements such as balance-of-system (BoS) units, cables, and auxiliary casing. These components are excluded based on prior studies that indicate their negligible contributions to overall environmental impacts [166,167]. 3.5. INVENTORIES FOR LARGE SCALES 55 The following component lifetimes are assumed: 30 years for PV installations [116], 12.5 years for inverters [46], 18 years for NMC batteries, and 19 years for LFP batteries [16]. To ensure consistency with the functional unit, the environmental impacts of each technology are normalized over its respective lifetime. Inverters and transformers generally have similar lifetimes, although certain parts may require replacement every 10 years [168]. Accordingly, a 12.5-year lifetime is adopted for inverters in this study [46]. 3.5.2 Inventories for Large-Scale Assessment Impact Model for Belgium Figure 3.7: Structure of the impact model (adapted from PAPER 6). Arrows represent electricity flows and the different colors represent the various electricity production and storage options (MV = medium voltage; LV = low-voltage; RSD = residential sector; COM = commercial sector; PV = photovoltaic; Back2Grid = electricity fed back to the grid; EV = electric vehicle). An ESM is employed to assess decentralized local electricity systems in Belgium, with a focus on the residential and commercial sectors [169,170]. Within each sector, electricity supply can originate from: • the low-voltage grid, • PV installations, and • battery storage systems. Excess electricity generated in either sector can be fed back into the grid (Back2Grid). If the grid’s capacity is insufficient to handle these outflows, electricity curtailment is applied (see Figure 3.7). 56 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES The model represents both direct interactions between the residential and commercial sectors via the low-voltage grid and indirect interactions through a shared EV fleet. It aims to capture the role of decentralized electricity generation, particularly PV installations combined with stationary batteries, and various EV charging strategies in shaping the Belgian energy system. The TIMES modeling framework is used to build the impact model [171]. In general, ESMs are linear programs that incorporate detailed technology specifications to minimize system costs through optimal capacity planning and dispatch. Model outputs include future projections of installed capacities, energy production, marginal commodity prices, and system-wide emissions [169]. The starting point is an existing model of the Belgian residential energy system, which is expanded to include decentralized generation and flexibility options [172]. The time resolution is hourly. To reduce computational burden, the model simulates three representative weeks: one in summer, one in winter, and one during the seasonal transition, thereby approximating a full year’s dynamics. The assessed years are 2030 and 2050, with 2014 serving as a reference case. Electricity demand profiles are derived from artificial load curves including household appliances and solar irradiation data. Further modeling details are available in PAPER 6. Scenario Definition Flexibility in electricity systems is introduced through battery storage and varying EV charging strategies. The capacity of PV installations and battery systems to meet demand is constrained by a prosumer potential (PP), which limits the share of demand met by locally generated and stored electricity. By default, this constraint is set to 50 % for all scenarios. Three modes of stationary battery integration are considered: 1. No battery storage, 2. Enabled stationary battery installations, 3. Additional interaction between stationary storage and the grid. These modes reflect uncertainties in the viability of battery business models and the potential for increased self-consumption or participation in energy arbitrage. In parallel, three EV charging schemes are introduced to evaluate demand-side flexibility: 1. Fixed charging: EV charging profiles and capacity are predefined. 2. Instant charging: EVs charge immediately upon arrival at home. 3. Flexible charging: EV charging is optimized to minimize system costs. By default, EVs are charged at home. In high-flexibility scenarios, workplace charging is also considered. While the EVs themselves are not included in the LCA, their electricity demand is incorporated to define the total system load. Combining battery storage modes and EV charging schemes, four scenarios are developed: 1. low_flex: Low flexibility, 2. med_flex: Medium flexibility, 3. high_flex: High flexibility, 4. high_flex_PPH: High flexibility with full prosumer potential (High Flex PPH). 3.6. APPLIED LIFE CYCLE IMPACT ASSESSMENT METHODS 63 Table 3.8: Applied midpoint LCIA categories and the abbreviations used throughout this thesis (Abbrev. = Abbreviation; CC = Climate change; HT = Human toxicity). EF v3.1 Midpoint ReCiPe 2016 Midpoint (H) V1.04 / World (2010) H Abbrev. Acidification Terrestrial acidification AC CC Global warming CC CC: biogenic – CCB CC: fossil – CCF CC: land use & land use change – CCL Ecotoxicity: freshwater Freshwater ecotoxicity ETF – Marine ecotoxicity ETM – Terrestrial ecotoxicity ETT Ecotoxicity: freshwater, inorganics – ETFI Ecotoxicity: freshwater, organics – ETFO Energy resources: non-renewable Fossil resource scarcity ERF Eutrophication: freshwater Freshwater eutrophication EUF Eutrophication: marine Marine eutrophication EUM Eutrophication: terrestrial – EUT HT: carcinogenic Human carcinogenic toxicity HTC HT: carcinogenic, inorganics – HTCI HT: carcinogenic, organics – HTCO HT: non-carcinogenic Human non-carcinogenic toxicity HTN HT: non-carcinogenic, inorganics – HTNI HT: non-carcinogenic, organics – HTNO Ionizing radiation: human health Ionizing radiation IR Land use Land use LU Material resources: metals/minerals Mineral resource scarcity ADP_MM Ozone depletion Stratospheric ozone depletion OP – Ozone formation, Human health OFH – Ozone formation, Terrestrial ecosystems OFT Particulate matter formation Fine particulate matter formation PM Photochemical oxidant formation: human health – POF Water use Water consumption WU 64 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES 3.7 Uncertainty and sensitivity analysis To assess the robustness of the developed LCA results, this thesis employs both uncertainty quantification and sensitivity analyses. These approaches address variability stemming from background data (e.g., inventory databases) and from assumptions in foreground system modeling. 3.7.1 Background Uncertainty Background uncertainty arises from the use of generic life cycle inventory (LCI) databases, such as ecoinvent. This type of uncertainty is typically addressed using Monte Carlo simulations with dependent sampling. Monte Carlo simulation is a probabilistic technique used to assess uncertainty in background data [187]. It consists of the following steps: First, probability distributions are assigned to the input data, nowadays often provided directly within the ecoinvent database. Next, values are randomly sampled from these distributions. The sampled data are then used in the LCA calculation to produce output results. In the fourth step, the sampling and calculation process is repeated N times to generate N output samples. As a result, a frequency distribution of the output data is obtained, approximating its probability distribution [187]. Standard statistical analyses can subsequently be applied for interpretation. In general, the larger the number of simulations (N), the better the approximation of the true probability distribution [187]. The following sections describe how uncertainty is addressed in each of the three system types. Wind Turbines For the wind turbine LCA model, 1,000 Monte Carlo iterations were performed. While this number does not guarantee full convergence of all output distributions, it follows the minimum recommended threshold in the literature for generating statistically representative results. Studies such as Igos et al. (2019) suggest that approximately 10,000 iterations are typically required to ensure convergence and stability across all indicators. However, given computational constraints and the exploratory nature of the current analysis, 1,000 iterations were considered a reasonable compromise between accuracy and efficiency [188]. These simulations evaluate uncertainty across both the IPCC 2021 CC midpoint category and all 25 EF v3.1 midpoint impact categories to develop a comprehensive uncertainty profile. The following indicators are extracted from the simulation results: • Mean environmental impact • Standard deviation • Coefficient of variation (CV) • 95 % confidence intervals (CI) • Shapiro-Wilk test p-values for assessing distribution normality These metrics help characterize the shape and spread of the output distributions and identify impact categories with high variability or non-normal behavior. EV Flexibility Services Due to the reduced modeling complexity in the LCI foreground of energy systems where flexibility services are provided, computational performance allows for 10,000 Monte Carlo iterations. The following statistical descriptors are used to evaluate the results: 3.7. UNCERTAINTY AND SENSITIVITY ANALYSIS 65 • Mean environmental impact • Standard deviation • CV • 95 % CI Additionally, minimum, maximum, and percentile values are considered. Given the large sample size and the need for detailed configuration comparisons, a discernibility analysis is applied. Discernibility analysis is a post-hoc, pairwise comparison method used to assess whether observed differences between simulation outcomes are meaningful in light of background uncertainty [189–191]. Unlike classical hypothesis testing, this approach does not rely on additional indicator calculation, such as p-values or F-statistics. Instead, the results from Monte Carlo simulations are compared iteratively against a baseline configuration. For each iteration, performance differences that fall within a predefined threshold (e.g., ±5or±50 gCO2eq/kWh, depending on system scale) are considered statistically indistinguishable. This allows for identifying configurations that consistently outperform others, as opposed to those whose advantages are masked by uncertainty. While variance-based methods such as analysis of variance (ANOVA) are common in classical statistics, they are less applicable in model-driven LCA contexts, where results often exhibit non-normal distributions and complex interdependencies. In contrast, pairing Monte Carlo simulations with discernibility analysis maintains the full distributional structure of the data and emphasizes practical rather than formal statistical significance. Second-Life Batteries and Macro-Level Assessment For the SLB case studies, background uncertainty could not be quantitatively assessed due to the absence of statistical distributions in the source data. Average values are used in place of probabilistic data, as the premise database did not support uncertainty information for the relevant background activities at the time of data extraction. Consequently, scenario analysis is employed to explore how background system variation affects CC impacts. Following this rationale, scenario analysis is also applied to the macrolevel use case, which considers system configurations labeled as reference,low_flex,medium_flex, and high_flex. 3.7.2 Foreground Uncertainty and Sensitivity Analysis Foreground uncertainty refers to assumptions made during system modeling and parameterization. This uncertainty is explored using both one-at-a-time sensitivity analysis and perturbation analyses, depending on system complexity. Most parameters are represented by a single value, reflecting specific circumstances at a given point in time. It is therefore essential to examine how such parameters influence the results of the model. Quantifying foreground uncertainty not only indicates whether the results are sensitive to particular parameters, but also reveals the extent and direction of this sensitivity. In extreme cases, alternative parameter choices may lead to substantially different results and potentially alter the overall conclusions of the study. One-at-a-Time Sensitivity Analysis For wind turbines, a one-at-a-time (OAT) sensitivity analysis is conducted for selected key parameters. In this approach, one parameter is varied while others are held constant, allowing direct attribution of output changes to the tested parameter. For wind turbines, two main parameters are examined: 66 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES 1. Turbine lifetime: 18 years (historical Danish median) and 30 years (advanced future scenario) compared to the default 20 years. 2. Grid connection type: comparison of electric power buses (default) versus transformer-based connection, which affects cabling and infrastructure impacts. Perturbation Analysis Perturbation analysis is used to assess how marginal changes in parameters influence the model output [189]. Key input parameters are modified by ±10 %, and sensitivity is expressed using a sensitivity factor (SF): SF = ∆R R0 ∆P P0 (3.11) Where: •∆R= change in result •R0= baseline result •∆P= change in parameter •P0= baseline parameter SFs typically range between -1 and 1. Values approaching or exceeding |0.8|indicate high influence and warrant prioritization in future model calibration. Applying perturbation analysis in the context of LCA is well established and already applied in previous studies [192,193]. For the modeling of flexibility services, a total of 12 parameters modified by ±10% are tested. Table 3.9 presents the mapping of microand mezzo-level parameters per sensitivity. Table 3.9: Mapping of sensitivities (SEN) to parameter modifications at microand mezzo-levels. Each parameter is perturbed by ±10% (PV = Photovoltaic; WT = Wind turbine; LFP = Lithium iron phosphate; NMC = Lithium nickel manganese cobalt oxide; EV = Electric vehicle; BESS = Battery electric stationary storage). Sensitivity Micro-level parameter Mezzo-level parameter SEN_1&2 PV lifetime PV lifetime SEN_3&4 Lifetime BESS WT lifetime SEN_5&6 Project lifetime Project lifetime SEN_7&8 EV lifetime Lifetime LFP SEN_9&10 EV battery capacity Lifetime NMC SEN_11&12 EV battery weight NMC density SEN_13&14 Energy density LFP density SEN_15&16 PV capacity PV capacity SEN_17&18 BESS size WT capacity SEN_19&20 Grid consumption LFP size SEN_21&22 Charged electricity to EV NMC size SEN_23&24 Discharged electricity from EV Grid consumption 3.7. UNCERTAINTY AND SENSITIVITY ANALYSIS 67 For the SLB use case, the perturbation analysis assesses only the lifetime in second life. Investigating other parameters is excluded as the lifetime is the value for which a high variation in literature is found, from around 4 up to 12 years (see Chapter 2). Also, the sensitivity of EoL treatment on the performance of the SLB LCA model is not further explored as its contribution in general is rather small and as SLB are expected to enter EoL only in 2045 due to the high level of uncertainty. For the macro-level scenario, seven parameters are tested: 1. Battery energy density 2. Residential battery capacity 3. Commercial battery capacity 4. Share of second-life batteries 5. Residential PV generation 6. Commercial PV generation 7. Solar irradiation To conclude this Section, parameters investigating the developed LCA model’s sensitivity to the uncertainty related to parameters applied in the foreground modeling are presented: 1. Energy density 2. Battery capacity (RES) 3. Battery capacity (COM) 4. Share of SLB 5. PV-generated electricity (RES) 6. PV-generated electricity (COM) 7. Solar irradiation 8. PV installation size The first four parameters directly target the modeling of batteries, whereas the last four parameters aim at understanding the sensitivity of parameters related to PV installations. Overall, these analyses identify parameters that most strongly influence CC impacts and provide insight into model robustness under plausible variations. 68 CHAPTER 3. METHODOLOGY FOR ASSESSING EMERGING ENERGY TECHNOLOGIES Chapter 4 Unveiling wind power’s environmental footprint SUMMARY | This Chapter presents an LCA model developed to assess the environmental impacts of onshore and offshore wind turbines with a particular focus on refining LCI with regional data. The model improves an existing LCA model by accounting for connection cable length to the electricity grid, calculating transportation distances between component manufacturing sites and installation locations, considering different foundation types based on sea depth, and utilizing country-specific datasets for installation, operation, and EoL treatment. The model is applied to a case study of one onshore and one offshore wind turbine in Belgium. Furthermore, a prospective life cycle assessment (PLCA) is conducted using two different premise databases to evaluate the future environmental impacts of potential wind turbines in Belgium. Additionally, a statistical analysis is performed to identify key parameters influencing the environmental performance of wind energy. The following conclusion can be drawn from the correlation, regression, and random forest analyses: 1. Geographical location strongly influences CC impact, but further validation is needed. Longitude emerges as the most influential predictor in both the regression model and random forest analysis, suggesting that transport distances, installation logistics, and grid infrastructure play a role in CC impact. Latitude also has a moderate effect, indicating that regional variations contribute to emissions. However, these findings partially contradict earlier results from Section 4.2, necessitating further investigation. 2. Larger turbines require more materials, but this is partially offset by higher energy yields. Hub height moderately affects CC impact due to higher material demand and more complex installation processes for taller turbines. However, higher AEP compensates for this increase, reducing emissions per kWh. Since turbines above 2 MW are modeled using scaled 2 MW LCI data, future research would benefit from more available data on such larger turbines. 3. Given that maintenance is modeled generally, it seems that a higher AEP correlates with lower maintenance CC impact. A negative correlation between AEP and maintenance CC impact suggests that more efficient turbines require fewer maintenance-related emissions per kWh. Obviously, this does not correspond to current practices and thus the maintenance modeling should be revised in the next model iterations. However, given the generalization in the maintenance modeling approach, further refinement of LCI data of maintenance is required to support this statement. 69 70 CHAPTER 4. UNVEILING WIND POWER’S ENVIRONMENTAL FOOTPRINT 4. Rated power and rotor diameter have a weaker influence on CC impact. Although rated power and rotor diameter are highly correlated, their direct influence on total CC impact remains weak. Random forest analysis confirms this, as these parameters rank among the least important features. This suggests that increasing turbine size does not necessarily result in higher environmental impact. This observation, which does not necessarily reflect findings of other studies, is again limited by the fact that the utilized LCI data are currently restricted to 2MW installations. 5. Disposal impacts remain largely independent of turbine size. Correlation results indicate that disposal-related emissions are not strongly influenced by most technical parameters, except hub height to some extent. This suggests that recycling and material recovery strategies should be prioritized in future wind turbine sustainability efforts, rather than focusing solely on design modifications. Overall, there is variability in the CC impacts of wind turbines. One key finding is that the components improved by incorporating regionalized LCI data have a relatively small effect on total CC impact compared to the main components such as tower, rotor, and foundations. However, differences across locations remain observable. Another insight is the importance of lifetime electricity production, which is consistently identified as the most influential parameter in the correlation, regression, and random forest analyses. This trend can be primarily linked to the selection of the functional unit (per kWh of electricity produced). Conversely, the influence of location parameters appears to be relatively minor, aligning with observations at the individual turbine level. From a macro perspective, offshore wind turbines demonstrate clear environmental advantages over onshore turbines. The case study shows that the CC impact of the analyzed offshore turbine is 8 gCO2eq/kWh, while 15 gCO2eq/kWh are calculated for the onshore turbine. This is consistent with national trends: the average CC impact of Belgium’s current offshore wind fleet is 5 gCO2eq/kWh, whereas the onshore fleet averages 8 gCO2eq/kWh. Looking ahead, the PLCA results indicate that the environmental performance of offshore wind will continue to improve, with future offshore turbines expected to achieve CC impacts below 4 gCO2eq/kWh by 2050. The primary driver of these impacts is component production, particularly for towers and foundations, whereas maintenance and transportation have relatively minor effects. The uncertainty and sensitivity analyses highlight the importance of turbine lifetime and grid connection choices in determining environmental performance. Extending the lifetime from 20 years to 30 years reduces CC impact, while variations in grid connection points have minimal influence. Additionally, Monte Carlo simulations confirm that uncertainties in background data do not substantially alter the overall conclusions of the LCA model. At the national level, the distribution of CC impacts in Belgium’s wind fleet further reinforces the environmental potential of offshore wind. While current onshore installations have a wider spread of CC impacts due to variability in AEP and turbine characteristics, offshore turbines consistently show lower emissions per kWh. Moreover, the future offshore fleet is projected to achieve CC impacts below 2 gCO2eq/kWh, further emphasizing its role in Belgium’s energy transition. Belgium’s wind energy sector is set to play a crucial role in the country’s shift toward a low-carbon electricity system. While current onshore installations are limited by lower CFs and aging installations, offshore wind turbines, both current and future, demonstrate higher efficiency and lower environmental impacts. The findings from this study underscore the importance of strategic planning and investment in offshore wind technology, as well as refining LCA methodologies to ensure more accurate sustainability assessments. The content of this chapter is strongly based on PAPER 1: D. Huber, M. Lavigne Philippot, L. E. Ramirez Camargo, M. Sierra Montoya, T. Coosemans, and M. Messagie, “Enhancing life cycle assessment with geographic data: spatially differentiated environmental impacts of European wind turbines,” submitted for publication to International Journal of Life Cycle Assessment, 2025. Cited as [28]. 4.1. USE CASE DESCRIPTION 71 4.1 Use case description Two different assessments are presented in this thesis. First, to show the functionality and output of the developed LCA model, a use case is defined, with an assessment of one onshore and one offshore turbine in Belgium. To align with the WIMBY interactive map, the rated power of the onshore and offshore turbine use cases are set to 3.4 MW and 10 MW respectively. For this assessment, the annual electricity production of the turbines is obtained from the WIMBY interactive map. The second evaluation seeks to assess the environmental impacts of the Belgian wind fleet. For the second assessment, the annual electricity production of the turbines is calculated. To support Belgium’s goal of climate neutrality, the country aims to expand its offshore wind capacity to 5.8 GW by 2030, 8 GW by 2040, and, based on exponential interpolation, 19 GW by 2050 [194, 195]. Since the LCA model assesses environmental impacts at the turbine level, estimating the rated power of future offshore turbines is necessary. The median rated power of the current offshore fleet is first determined, and a normal distribution is fitted to the data. By 2050, the median is set to 15 MW, reflecting the largest turbines available today [47]. This shift is applied to the fitted normal distribution while maintaining the original variance, resulting in a projected distribution of rated power for future offshore turbines (see Figure 4.1). Turbine capacities are then randomly sampled from this distribution and rounded to the nearest hundred kW until the total installed capacity is met. Hub height and rotor diameter are estimated using scaling models from Sacchi et al. (2019) [66]. Figure 4.1: Rated power distribution of the current and future Belgian offshore wind fleet. Turbines are randomly assigned locations within the Prince Elisabeth Zone in the Belgian North Sea, respecting a minimum spacing of four rotor diameters. While this placement is a simplification, detailed spatial planning is beyond the scope of this study. To allocate grid connection impacts, turbines are grouped into six parks of varying sizes, totaling 1,150 turbines. This parameter is simply used to allocate the impact of grid connection between the turbines within the same park. Future electricity production calculation involves uncertainty due to factors such as climate-driven wind speed variations 72 CHAPTER 4. UNVEILING WIND POWER’S ENVIRONMENTAL FOOTPRINT and technological advancements in power curves. Instead of using wind speed and power curves, the AEP is estimated based on projected CFs. Li et al. (2022) predict a global offshore CF increase from 50 % in 2020 to 60 % in 2040, suggesting a 70 % CF by 2050 via exponential interpolation [56]. A normal distribution derived from current CF data is used to project future CF values while ensuring they remain between 0 and 1 (see Figure C.1). These projected CFs are then randomly assigned to future turbines, enabling the estimation of annual electricity production (AEP) by multiplying each turbine’s rated power by its assigned CF and annual hours. The 1,217 turbines installed in Belgium up until the end of 2021 yield 13.4 TWh annually, with almost three-quarters of this generation coming from the offshore wind fleet. In the context of Belgium’s final electricity consumption of 423.4 TWh in 2023, the wind fleet contributes about 3 % to the total electricity consumption [196]. The AEP per rated power of the onshore and offshore Belgian wind fleet is illustrated in Figure 4.2. Figure 4.2: Annual electricity production per rated power of the Belgium onand offshore wind turbines (AEP = annual electricity production). One observation is that the AEP increases with higher rated power. However, when examining the AEP of the onshore fleet (top subplot in Figure 4.2), turbines with rated powers of 5, 6.5, and 8 MW have 4.4. MACRO-LEVEL EVALUATION 79 4.4 Macro-level evaluation Figure 4.6 shows the results of the calculated CC impact of the current and future Belgian wind fleet, summarized in a histogram. Figure 4.6: Climate Change Impacts of onand offshore wind turbines in Belgium (IPCC = Intergovernmental Panel on Climate Change; GWP = Global warming potential. The first observation is that there are considerably fewer offshore turbines installed than onshore (816 onshore vs. 401 offshore). The corresponding mean values of the current onand offshore and the future offshore fleet are 8, 5 and 2 gCO2eq/kWh respectively. After removing 31 onshore turbines with a CC impact higher than 20 gCO2eq/kWh, the represented data of the current fleet represents 1,186 turbines. The reason for the high CC impacts of those outliers is a low AEP compared to the other onshore wind turbines. Apart from the distribution of the CC, Figure 4.6 points out another finding very clearly: the environmental potential wind energy has. While the CC impact distribution of the current onshore wind fleet is averaging around 8 gCO2eq/kWh, the main CC impact distribution of the current offshore wind fleet is around 4 gCO2eq/kWh. Lastly, the CC impact distribution of the future offshore fleet is divided into two groups: most of the future offshore turbines have a CC impact below 2 gCO2eq/kWh, while the CC impacts of a second group of future offshore turbines are slightly higher, but still below the CC impacts of the current wind fleet. Therefore, Figure 4.6 can be interpreted as an environmental advocate for future wind turbine deployment. After getting a general understanding of the CC impacts and its distribution in the current onand offshore and the future Belgian offshore wind fleet, the next paragraphs will explore the structure of the LCA model and identify the model’s sensitivity towards certain parameters. Figure 4.7 presents scatter plots of total CC impact of the onshore fleet against CF, cable length and transport distances. Total CC impacts plotted against other parameters, such as rated power, hub height, rotor diameter, sea depth, longitude, latitude along with CC impacts of the offshore fleet are visualized in Figure C.5 to Figure C.9. In general, the presented Figures point out very well the wide spread of CC impacts of onshore wind turbines when plotted against the technical parameters. In contrast, 80 CHAPTER 4. UNVEILING WIND POWER’S ENVIRONMENTAL FOOTPRINT the CC impacts of offshore turbines are more clustered and condensed. While the scatter plot allows to visually identify first trends, e.g. a negative correlation between CF and CC impacts for onand offshore turbines, other relations are less obvious and call for further assessment. The visual trends suggest potential relationships between CC impact and these variables, particularly a positive trend with hub height and a noticeable spatial dependency with longitude and latitude. However, to reveal the full strength and nature of these relationships, further statistical assessments are required and to start with, results from the correlation analysis are presented. Figure 4.7: Climate Change impacts of the onshore wind fleet plotted versus capacity factors, cable lengths and transport distances. Further visualization for rated power, hub height, rotor diameter, sea depth and longand latitude for onshore are provided in the Appendix C along with the same visualization for CC impacts of the offshore fleet. 4.4. MACRO-LEVEL EVALUATION 81 4.4.1 Correlation analysis Figure 4.8 visualizes the correlation between technical parameters (rated power, hub height, rotor diameter, longitude, latitude, and AEP) and the CC impacts of the current Belgian wind fleet. The future offshore wind fleet is excluded from this assessment because the LCA model was originally designed for smaller wind installations and does not include the updated LCI data for large offshore turbines. Additionally, AEP for the future offshore fleet was approximated based on rated power and annual operating hours, making this type of analysis highly predictive rather than empirical. Nevertheless, the correlation matrix, the regression model summary and the random forest model of the future Belgian offshore wind turbines are included in Figures C.10, C.12, C.15. Figure 4.8: Correlation factors between technical parameters and climate change impacts of the current Belgian wind fleet. ’P_rated_kW’ to ’aep_kwh’ are technical parameters, ’Component production’ to ’Total’ are CC impacts. Focusing on the existing Belgian wind fleet, the first observation is that no exceptionally strong correlations exist between technical parameters and CC impact categories (component production, assembly, transport, maintenance, disposal, and total). The strongest correlation with total CC impact is observed for longitude and hub height (both below 0.6), followed by latitude and AEP, while rated power and rotor diameter show a weaker influence. Longitude and latitude have moderate correlations (0.58 and -0.45, respectively) with total CC impact. This suggests that geographical location influences CC impact, likely due to regionalized factors such as sea depth, cable length, and transportation distances, contradicting the findings from Section 4.2. Offshore turbines tend to have higher transport and installation emissions than onshore turbines, affecting the total CC. Onshore turbines in different regions may also be subject to longer transportation routes, leading to differences in the transport stage. A higher hub height is associated with a greater total 82 CHAPTER 4. UNVEILING WIND POWER’S ENVIRONMENTAL FOOTPRINT CC impact. This is not surprising, as taller turbines require larger tower sections, increased material use, and more complex installation processes. However, a taller hub height also yields more electricity (AEP), which helps reduce gCO2eq/kWh, mitigating some of the initial impact. AEP has a medium correlation with total CC impact but shows a strong negative correlation with maintenance CC impacts. This could imply that higher-energy-producing turbines require less maintenance per unit of electricity produced, likely due to more efficient, newer designs. However, the AEP-CC impacts correlation is not dominant, meaning other factors like material choice and transport distances are more influential. Rated power and rotor diameter are highly correlated (0.93), which is logical, as larger rotors are needed for higher-capacity turbines. However, their correlation with total CC impacts is weaker, suggesting that simply increasing rated power does not drastically increase emissions per kWh. Since the input data is limited to turbines up to 2 MW, the model scales up larger turbines based on 2 MW LCI data, meaning that these findings may not be fully representative of larger installations. Maintenance CC impact is most strongly (negatively) correlated with AEP and the rotor diameter. This would suggest that larger, more productive turbines require less maintenance per kWh. Disposal impacts show relatively uniform correlations across technical parameters, apart from hub height, meaning disposal impacts are largely independent of turbine size but may be more affected by material composition. Thus, the correlation matrix can be summarized as follows: First, regional differences in CC impact exist but require further validation. Wind turbines in different regions exhibit variations in CC impact, likely due to logistical factors such as transport distances, installation conditions, and grid connection requirements. However, these findings partially contradict those from Section 4.2, suggesting that the true impact of geographical factors requires further investigation. Second, larger turbines require more material but may compensate through higher energy yields. Taller turbines (with higher hub heights and larger rotors) require more materials and complex installation processes, increasing component production emissions. However, this effect is partially offset by higher energy production (AEP), which can reduce emissions per kWh. Since the input data is limited to turbines up to 2 MW, this statement is primarily applicable to smaller turbines. For larger turbines exceeding 2 MW, LCI data was scaled up from 2 MW installations, meaning these results should be interpreted with caution. Third, higher AEP correlates with lower maintenance CC impact, but modeling limitations exist. Wind turbines with higher energy output (AEP) tend to have lower maintenance emissions per kWh, which could indicate greater efficiency in newer turbine designs. However, since the maintenance modeling approach is relatively general, this trend should not be overstated and further refinement of maintenancerelated LCI data would be beneficial. Fourth, the relationship between rated power and CC impact is not straightforward. Rated power and rotor diameter are highly correlated (0.93), which is expected due to design constraints. However, their correlation with total CC impact is weaker, suggesting that simply increasing turbine capacity does not necessarily lead to a proportional increase in emissions per kWh. Given the simplified scaling approach for turbines exceeding 2 MW, further LCI data is needed to assess the true environmental impact of large turbines. Fifth, the disposal impacts remain largely independent of turbine size. Correlations suggest that disposal-related emissions are not strongly influenced by most technical parameters, except hub height to some extent. This implies that recycling and material recovery strategies should be prioritized in future wind turbine sustainability efforts, rather than focusing solely on size-based design changes. 4.4.2 Ordinary least squares regression model To further investigate the relationships observed in the correlation matrix, an Ordinary Least Squares regression model is applied to assess how technical parameters influence the total CC impact of the Belgian wind fleet. The results indicate that 56.1% of the variation in total CC impact is explained by the 4.4. MACRO-LEVEL EVALUATION 83 selected predictors (R2= 0.561) (Figure C.11) . The model is statistically significant (F-statistic = 251.1, p< 0.001), suggesting that the included parameters provide meaningful insights into the determinants of CC impact. The main findings of the regression model can be summarized as follows: First, hub height shows the strongest positive association with total CC impact (β= 0.0796, p< 0.001), indicating that taller turbines contribute more to CC impact, likely due to higher material demand and higher assembly efforts. Second, longitude and latitude are also important predictors (β= 0.5781, p< 0.001 and β= 0.3911, p= 0.006, respectively). This suggests that regional location influences CC impact, possibly due to differences in component transportation, grid infrastructure, or considering local assembly. Third, annual electricity production (AEP) has a strong negative effect on total CC impact (β= -1.045e-07, p< 0.001), reinforcing the observation that higher electricity generation helps reduce emissions per kWh. Fourth, rated power and rotor diameter do not have statistically significant effects on total CC impact when considered independently (p= 0.362 and p= 0.437, respectively), indicating that increasing turbine capacity alone does not necessarily lead to higher environmental impact. To account for potential dependencies between technical parameters, a second regression model is tested that included interaction terms (e.g., rated power × rotor diameter, hub height × diameter, longitude × latitude). The model’s output is presented in Figure C.13. The inclusion of these terms increased the explained variance from 56.1% (R2= 0.561) to 64.2% (R2= 0.642), indicating that interactions help capture additional variation in CC impacts. In particular, the interaction between rated power and rotor diameter is significant (β= 2.65e–05, p< 0.001), while the interaction between diameter and hub height also contributes (β= –0.0013, p< 0.001), highlighting that the effect of turbine size parameters on CC impact cannot be fully understood when considered independently. By contrast, rated power and rotor diameter alone are not significant in the baseline model (p= 0.362 and p= 0.437, respectively), which suggests that a greater influence can be observed when assessed combined. In the second regression model, coefficients for some predictors, such as latitude (β= 0.3911, p= 0.006 in the baseline vs. β= 0.978, p< 0.001 in the interaction model), shifted compared to the baseline model, reflecting the redistribution of explained variance between main effects and their combinations. Although the interaction model provides a better statistical fit, its interpretation is less straightforward due to multicollinearity (condition number 6.71e+07) and changing coefficient meanings. Therefore, the baseline model is retained as the main analytical tool for clarity, with the interaction model serving as a robustness check confirming the interdependency of design parameters. In general, the high condition number (2.49e+09 in the first and 6.71e+07 in the second regression model) suggests potential multicollinearity issues, meaning that some predictor variables may be strongly correlated with each other, potentially affecting the model’s stability. Further investigations, such as a random forest model, are applied to address these concerns. 4.4.3 Random forest model To complement the regression analysis, a random forest model is applied to assess the relative importance of technical parameters in predicting the CC impact of the current Belgian wind fleet. Unlike regression, which assumes a linear relationship between predictors and the dependent variable, random forests are a non-parametric method that can capture complex interactions and nonlinear effects between features. The feature importance values indicate how much each predictor contributes to reducing the model’s error. The results reveal that longitude has the highest feature importance ( 0.62), reinforcing its strong influence on CC impact, as also observed in the regression analysis. This suggests that geographical factors may be key determinants of CC impact (Figure 4.9). 84 CHAPTER 4. UNVEILING WIND POWER’S ENVIRONMENTAL FOOTPRINT Figure 4.9: Feature importance of predictors in the Random Forest model, showing the relative contribution of each variable to the model’s predictive power for the current Belgian wind fleet (units: dimensionless). Latitude follows with a moderate importance ( 0.15), further highlighting the spatial variability in CC impact across different locations. The AEP is also an important predictor ( 0.10), consistent with previous findings that higher energy generation helps reduce emissions when normalized per kWh. Hub height ( 0.05), rated power ( 0.03), and rotor diameter ( 0.02) have lower feature importance values, suggesting that these design parameters play a smaller role compared to geographical factors. The random forest results provide an important complement to the regression model, as they confirm the dominance of location-based parameters (longitude and latitude) over purely technical parameters like rotor diameter and rated power. This suggests that site-specific conditions, rather than just turbine design, are crucial in determining CC impact. 4.5 Discussion and limitations In the previous sections, the strength of the developed model is demonstrated, supported by statistical analysis to identify important input data and parameters. One point to highlight concerns the case studies, which do not represent the average or default condition of the entire country. Instead, the results of the case studies represent a very specific turbine setup at a given location and thus cannot be understood as a country representation of the entire wind fleet. A good example is the CF of offshore wind turbines that are selected as case studies. Those values originate from the interactive WIMBY map and are calculated using a detailed model developed by the Technical University of Denmark. With a CF of 56 %, it exceeds the average CF of the EU offshore wind fleet in 2023 by around 22 % provided by Wind Europe [145]. As shown in the Section 4.4, CFs are found to be a major parameter when determining CC impacts of wind turbines at kWh-level. Thus, the presented results of the case studies have to be considered cautiously given those high CFs. The model and this work are subject to certain limitations. Following the previous point about the CF of the case studies, the calculation of the electricity production of the EU wind fleet is subject to simplifications. Instead of considering turbine-specific cut-in and cut-off speeds, affecting the determined power curves and thus the annual electricity production, the same cut-in and cut-off wind speeds for all the EU wind turbines are applied, as recommended by Saint-Drenan et al. (2020) [142]. Furthermore, 4.5. DISCUSSION AND LIMITATIONS 85 the same parametric power curve model, based on rated power and rotor diameter, is used for all the EU wind fleet, neglecting for example turbulence kinetic energy. Ignoring turbine-specific conditions in the power curves such as cut-in and cut-off wind speeds or turbulence kinetic energy have a direct effect on the subsequently calculated electricity production and thus on the environmental impacts of the Belgian wind fleet. To decrease the uncertainty in the electricity production calculation, one way could have been to validate the determined power curves with a commercial database. This, however, would require a matching between the Belgian wind fleet turbines and the particular wind turbine model. As a consequence, the validation of the calculated power curves remains unaddressed. As the main objective of this research is to develop a regional LCA model for the calculation of onand offshore turbines and not to determine the electricity yield of the existing Belgian wind fleet, efforts are limited to this simplified approach, knowing that this is not a precise calculation with the objective to reflect real-world production. Another limitation concerns input data, affecting all life cycle stages: over the last years, larger turbines entered the market with new generator technologies. Such new generator technologies not only vary in material quantity, but also in material composition such as permanent magnets, being known for containing critical raw materials. The current model is in general capable of scaling up the turbine based on larger rated power, but does not take into account, for example, new generator technologies. To the author’s knowledge, no study exists yet containing such detailed information as on material quantity per sub-component level, which is the required input data for the current LCA model. Therefore, the first claim is towards industry to be more transparent in their reporting, which would allow the scientific community to increase the accuracy of their work and thus credibility of their results. In a similar direction goes the need for more data regarding EoL treatment. To put things into context: based on the LCI model, an approximation of materials currently bound in the Belgian wind fleet can be made. The current Belgian wind fleet binds about 1,100 kilotonnes (kt) of metals, 84 kt of composites, and 3 kt of plastics. To provide context for these Figures: 1. If the total metal content were composed entirely of steel, it would be sufficient to construct 150 replicas of the Eiffel Tower, which contains approximately 7.3 kt of wrought iron. 2. The amount of composite materials in the fleet is equivalent to the takeoff weight of approximately 560 Airbus A320 aircraft, each weighing around 75 tonnes at full capacity. 3. The volume of plastics used corresponds to the weight of approximately 200 million 0.5-liter PET bottles, assuming an average bottle weight of 15 grams. These comparisons highlight the substantial material demand embedded in the current wind fleet. With the ambition to increase the share of RET in electricity generation, the demand for materials will continue to rise. Consequently, the materials currently bound in wind turbines should be considered a valuable resource rather than mere waste. When analyzing the ecosystem of EoL treatment of wind turbines, it becomes very quickly evident that all efforts on treatment of current retired wind turbines are, if they exist at all, at lab scale. Thus, reliable data to model a future EoL are not available. Including any other EoL treatment would have been very hypothetical, as currently, no one knows how large scale future treatment of retired wind turbines would look like and thus increased uncertainty. The problems related to EoL treatment of wind turbines are addressed in the first WIMBY deliverable [34]. While metals can achieve high recycling rates—up to 95 % for certain materials—the recycling of composite materials remains a major challenge. As new wind turbines are continuously installed and older ones approach the end of their operational lifetimes, the lack of economically and technologically feasible processing methods for used components, particularly wind turbine blades, presents a critical issue. The development of scalable, cost-effective recycling technologies for various composite materials remains an open area of research. Further details about the EoL treatment are provided in the first deliverable of the WIMBY project entitled ’LCA results report and spatially explicit data (a)’ [34]. As the current Belgian wind fleet ages, it should 86 CHAPTER 4. UNVEILING WIND POWER’S ENVIRONMENTAL FOOTPRINT be up to policy makers to start thinking about a legal framework for those huge upcoming waste streams. Researchers can support this process by starting to model and assess those waste streams, for example using a material-flow analysis combined with life cycle assessment. However, this task is out of the thesis scope and will remain open for future research. A very nice example on individual turbine level is the paper of Diez-Cañamero and Mendoza (2023), where they analyze the circular economy performance and carbon footprint of wind turbine blades [197]. Next, some simplifications are made when modeling the regional LCI: first, for both transportation distances and cable lengths, the direct distance is considered between the manufacturing and turbine location for transportation and the buses and the turbine location for cable lengths. Obviously, this does not depict reality, e.g. as for transportation, the trucks would travel on the road and use existing infrastructure. Similarly, the cable will not necessarily be able to cross all different areas, e.g. lakes, high mountains, buildings, etc. Nevertheless, this input data is used in the LCA model as the contribution towards the final results remain so little, that changing those input data would affect total environmental impacts only marginally. Until then, this model can still be used as a valid proxy. The next point is rather an open call for both manufacturers and operators. As the analysis showed, the environmental impacts decrease with longer lifetimes. This can be already considered in the design phase, e.g. the lifetime needs to be extended as much as possible, making turbines ready for operation as long as possible. Of course, guaranteeing long lifespans is not purely the responsibility of the manufacturer, but also requires regular maintenance efforts from the operator. Figure 4.10: Climate change impacts of the current and future Belgian wind fleet, overlaid with the Natura 2000 protected areas on the Belgian map. Future turbines are modeled only offshore and are depicted in dark blue/black dots on the two upper right areas. 4.5. DISCUSSION AND LIMITATIONS 87 Finally, only new offshore installations are considered in the context of Belgium’s future wind energy development. However, when plotting only the CC impact of the turbines on a map, it appears that there is still considerable space available for onshore wind turbines, particularly in the south of Belgium. Overlaying the Natura 2000 map onto the CC impact map of Belgian turbines provides an initial indication of spatial limitations for further wind power deployment (see Figure 4.10). Natura 2000 covers protected areas for some of the most valuable and threatened species and habitats across Europe [198]. Beyond protected areas, numerous other constraints restrict land availability for wind energy deployment, including existing infrastructure, buildings, and military zones. A first approximation of current land use in Belgium is provided by the CORINE Land Cover dataset from the Copernicus Institute. Figure C.16 in Appendix B illustrates Belgium’s land use types, ranging from broad-leaved forests (depicted in light green) to continuous urban fabric (depicted in red) [199]. As discussed in Section 4.1, most of Belgium’s current onshore wind turbines have a rated power between 2 and 2.5 MW, with an average age of around 12 years, highlighting their technological obsolescence. Given Belgium’s limited land availability, the aging onshore fleet presents an opportunity for repowering, where older turbines are replaced with newer, larger, and more efficient models. However, repowering of existing onshore turbines is beyond the scope of this study and remains an avenue for future research. 88 CHAPTER 4. UNVEILING WIND POWER’S ENVIRONMENTAL FOOTPRINT 5.2. MICRO-LEVEL RESULTS 95 5.2 Micro-level results 5.2.1 Contribution analysis The CC impact per kWh of electricity generated at the micro level is highest for the configuration using uni-directional charging without BESS, reaching 165 gCO2eq/kWh (Figure 5.4). Introducing smart charging reduces CC impacts by over 22 %. Across all charging strategies, grid electricity remains the main contributor, accounting for 76-90 % of total CC impacts. The second-largest contributor is PV installations, though their contribution is substantially lower. In the system with BESS, PV installations contribute up to 13 gCO2eq/kWh. The impacts from BESS and EV batteries together represent only 46 % of total CC impacts. Figure 5.4: Contribution to climate change impact of assets at a micro-level electricity system in Belgium (PV = Photovoltaic installations; EV = Allocated impact of manufacturing a battery for an electric vehicle; BESS = battery electric stationary storage). Despite the added emissions from battery production, installing a BESS leads to overall reductions in CC impacts due to decreased reliance on grid electricity. The lowest CC impact (65 gCO2eq/kWh) occurs in the configuration combining smart charging with BESS. This represents a 60 % reduction compared to the uni-directional, no-BESS baseline. For context, the average CC impact of the Belgian electricity grid 96 CHAPTER 5. ASSESSING FLEXIBILITY SOLUTIONS IN ENERGY SYSTEMS is 153 gCO2eq/kWh, indicating that a PV-coupled system with uni-directional charging performs worse than the existing grid. In contrast, combining smart charging and BESS can lower impacts by up to 61 %, highlighting the environmental advantage of flexible, storage-supported configurations. 5.2.2 Uncertainty evaluation Global uncertainty assessment For the Monte Carlo simulation, the remodeled hourly grid mix was substituted with the ecoinvent dataset market for electricity, low voltage for Belgium, enabling the inclusion of uncertainty data. The results confirm the trend observed in the static LCA: systems with uni-directional charging and no BESS exhibit the highest mean CC impact (230 gCO2eq/kWh). Adding BESS lowers this to 104 gCO2eq/kWh, the lowest among all configurations. Switching to smart charging reduces the impact to 178 gCO2eq/kWh (without BESS) and 146 gCO2eq/kWh (with BESS). Only the uni-directional, no-BESS configuration exceeds the mean CC impact of the current Belgian electricity mix (206 gCO2eq/kWh); all others perform better. Standard deviations range from 5 to 12 gCO2eq/kWh, with Uni - no BESS showing the highest variability (SD = 12), and Uni - BESS the lowest (SD = 5). Despite these standard deviations, the 95 % confidence intervals remain narrow, typically within ±1 gCO2eq/kWh (Figure 5.5). Figure 5.5: Results of a Monte Carlo simulation for a micro-level electricity system in Belgium (n= 10,000). Extreme outliers were removed using the interquartile range method to improve statistical robustness (BESS = Battery electric stationary storage; SD = Standard deviation; CI = Confidence interval). The Monte Carlo mean values are 25-50 gCO2eq/kWh higher than the corresponding static LCA values. While this difference could stem from model structure, a more likely explanation lies in the electricity dataset: the static model uses hourly values, whereas an average ecoinvent dataset is included 5.2. MICRO-LEVEL RESULTS 97 in the Monte Carlo simulation, which differs by 50 gCO2eq/kWh. Given the dominance of grid electricity in micro-level systems, this substitution likely explains the divergence. Additionally, normalization per kWh compresses the output variance, resulting in artificially narrow confidence intervals. This effect is clarified by comparing normalized and unnormalized distributions (Figure C.22), which show a broader spread in absolute (unnormalized) terms. Because the distributions of different configurations partly overlap, their environmental performance cannot be fully distinguished in the Monte Carlo plot in Figure 5.5. To address this, a discernibility analysis is conducted, using a threshold of ±50 gCO2eq/kWh to assess whether performance differences between configurations are statistically meaningful (see Figure 5.6). Figure 5.6: Results of the discernibility analysis in the macro-level use case for climate change impacts (EF v3.1, midpoint, CC) comparing three energy systems with different charging strategies against the default configuration (uni-directional charging without BESS). Each dot represents the outcome of a single Monte Carlo run (n = 10,000). Points falling within the grey zone indicate statistically similar results (±50 gCO2eq/kWh); points above or below indicate which scenario is environmentally preferable. 98 CHAPTER 5. ASSESSING FLEXIBILITY SOLUTIONS IN ENERGY SYSTEMS The results visualized in Figure 5.6 are quantified in Table 5.3. Table 5.3: Quantification of discernibility results of the micro-level use case. Results are classified as similar when they lie within a range of ±50 gCO2eq/kWh compared to the baseline scenario. Baseline Compared Scenario Better than baseline (%) Similar (%) Worse than baseline (%) Uni – no BESS Uni – BESS 99.90 0.00 0.00 Smart – no BESS 69.00 31.00 0.00 Smart – BESS 99.90 0.10 0.00 Uni – BESS Uni – no BESS 0.00 0.00 99.90 Smart – no BESS 0.00 0.10 99.90 Smart – BESS 0.00 99.10 0.90 Smart – no BESS Uni – no BESS 0.00 31.00 69.00 Uni – BESS 99.90 0.10 0.00 Smart – BESS 0.10 99.90 0.00 Smart – BESS Uni – no BESS 0.00 0.10 99.90 Uni – BESS 0.90 99.10 0.00 Smart – no BESS 0.00 99.90 0.10 1. Uni - no BESS as baseline. Relative to Uni - no BESS, both configurations considering storage, Uni - BESS and Smart - BESS, indicate lower CC impacts in nearly all iterations (99.90%), with no instances where CC impacts are higher and at most 0.10% of iterations lead to similar impacts. Smart - no BESS also improves over the baseline in most cases (69.00%), with the remaining 31.00% falling within the discernibility band. Across all iterations, the baseline, Uni - no BESS, is never identified as the configuration with the lowest CC impacts. 2. Uni - BESS as baseline. Against Uni - BESS, both Uni - no BESS and Smart - no BESS result in higher CC impacts in almost all iterations (99.90%), confirming the positive effect of configurations with stationary storage over configurations without stationary storage. CC impacts of Smart - BESS indicate practical equivalence: 99.10% of iterations result in similar CC impacts, with only 0.90% of all iterations in which Uni - BESS indicates slightly lower CC impacts (i.e., Smart - BESS is worse in 0.90% of all iterations). 3. Smart - no BESS as baseline. When Smart - no BESS serves as baseline, Uni - BESS has lower CC impacts in 99.90% of iterations, underscoring the decisive benefit of storage regardless of the charging strategy. The comparison with Smart - BESS yields similar CC impacts in almost all iterations (99.90%), with only 0.10% favoring Smart - BESS. Versus Uni - no BESS, the baseline has lower CC impacts in 69.00% and similar CC impacts in 31.00% of iterations, reflecting a consistent but more modest benefit from smart operation alone. 4. Smart - BESS as baseline. Relative to Smart - BESS,Uni - no BESS almost always yields higher CC impacts (99.90%) and is rarely similar (0.10%). Smart - no BESS is effectively indistinguishable (99.90% similar; 0.10% worse), indicating that smart charging without storage does not make a particularly pronounced 5.2. MICRO-LEVEL RESULTS 99 difference in CC impacts. Compared with Uni - BESS, CC impacts are again equivalent (99.10% similar), with a slight 0.90% edge in favor of Uni - BESS. In summary, across pairwise contrasts within the ±50 gCO2eq/kWh discernibility band, introducing stationary storage in a system is found to have a much higher potential to reduce CC impacts: both Uni - BESS and Smart - BESS overwhelmingly outperform Unino BESS, while Smart - no BESS offers moderate improvements over Uni - no BESS but remains inferior to any BESS configuration. CC impact differences between the two BESS options are negligible (about 99% similar), with only small improvements appearing in isolated iterations. Overall, integrating BESS yields robust performance gains, whereas introducing smart charging is not identified as having a major reduction potential for CC impacts. Sensitivity assessment Figure 5.7: Results of the perturbation analysis for the LCA model of a micro-level electricity system in Belgium (BAU = smart_BESS; SEN_1&2 = PV lifetime; SEN_3&4 = Lifetime BESS; SEN_5&6 = Project lifetime; SEN_7&8 = EV lifetime; SEN_9&10 = EV battery capacity; SEN_11&12 = EV battery weight; SEN_13&14 = Energy density; SEN_15&16 = PV capacity; SEN_17&18 = BESS size; SEN_19&20 = Grid consumption; SEN_21&22 = Charged electricity to EV; SEN_23&24 = Discharged electricity from EV). The perturbation analysis is conducted on the configuration with smart charging and BESS, as it includes all relevant parameters. Following Heijungs and Kleijn (2001), parameters with a SF above 0.8 are considered highly influential [189]. As visible in Figure 5.7, only three cases meet this criterion: SEN_5 and SEN_6 (system lifetime) and SEN_19 (grid consumption). These parameters affect the calculated functional unit (lifetime electricity output) and are expected to show high sensitivity. While lifetime is fixed across configurations, grid electricity varies depending on charging strategy. All other parameters, including PV/BESS lifetimes, capacities, and EV battery characteristics, yield SF values below 0.2. This indicates that the model is generally robust to most technical assumptions, with lifetime and grid reliance being the most critical inputs. 100 CHAPTER 5. ASSESSING FLEXIBILITY SOLUTIONS IN ENERGY SYSTEMS 5.2.3 Other impact categories Using the ecoinvent electricity dataset instead of hourly grid data leads to a shift in the ranking of configurations. Unlike the CC impact results using hourly mixes, where Smart - BESS performed best, the modified model identifies Uni - BESS as the lowest-impact configuration across all EF v3.1 midpoint categories (Figure 5.8). Smart - no BESS and Smart - BESS rank second and third worst, respectively. Despite variation in absolute values, this relative ranking remains consistent across categories. Figure 5.8: Results of all midpoint impact categories from EF v3.1 for the LCA model of a micro-level energy system with different charging strategies in Belgium. Cell labels show each configuration’s impact relative to the worst case within the same row (100% = worst/highest impact; smaller percentages indicate lower impacts, e.g., 60% means a 40% reduction vs. the worst). Values in cells are dimensionless (%). (CC = Climate change; CCB = Climate change: biogenic; CCF = Climate change: fossil; CCL = Climate change: land use and land use change; ETF = Ecotoxicity: freshwater; ERR = Energy resources: non-renewable; EUF = Eutrophication: freshwater; EUM = Eutrophication: marine; EUT = Eutrophication: terrestrial; HTC = Human Toxicity: Carcinogenic; HTCI = Human Toxicity: Carcinogenic, Inorganics; HTCO = Human Toxicity: Carcinogenic, Organics; HTN = Human Toxicity: Non-carcinogenic; HTNO = Human Toxicity: Non-carcinogenic, Organics; HTNI = Human Toxicity: Non-carcinogenic, Inorganics; IR = Ionising radiation: human health; LU = Land use; ADP_MM = Material Resources: Metals/Minerals; OD = Ozone Depletion; PM = Particulate Matter Formation; POF = Photochemical Ozone Formation: Human Health; WU = Water use). Grid electricity remains the dominant contributor in all scenarios, followed by PV installations (see Figure C.24). In ETF, PV systems contribute over 50 % of total impacts, regardless of configuration. BESS contributions remain moderate, reaching up to 20 % in select categories. These results confirm that 5.3. MEZZO-LEVEL RESULTS 101 smart charging and BESS offer benefits across multiple impact categories. However, they also highlight how the choice of background dataset can affect the relative ranking of system configurations. While Smart - BESS performs best in models using hourly grid data, Uni - BESS emerges as the top performer under the average mix assumptions. 5.3 Mezzo-level results 5.3.1 Contribution analysis Compared to the micro-level use case, CC impacts are lower across all system configurations at the mezzo level. For instance, uni-directional charging without BESS results in 41 gCO2eq/kWh, compared to 165 gCO2eq/kWh at the micro level. Additionally, the variation in CC impacts across configurations is relatively small, ranging from 32 to 41 gCO2eq/kWh, indicating less performance variability (see Figure 5.9). Figure 5.9: Contribution to climate change impact of assets at a mezzo-level electricity system in Belgium (PV = photovoltaic installations; EV = allocated impact of manufacturing a battery for an electric vehicle; LFP = lithium iron phosphate; NMC = lithium nickel manganese cobalt oxide). 102 CHAPTER 5. ASSESSING FLEXIBILITY SOLUTIONS IN ENERGY SYSTEMS Among the configurations, smart charging with BESS performs best, achieving a 21 % reduction in CC impacts relative to the baseline (uni-directional without BESS). Smart charging without BESS and uni-directional charging with BESS both result in 36 gCO2eq/kWh, representing 13 % and 12 % reductions respectively. While BESS integration introduces additional emissions from battery production, its impact remains limited. An analysis of the contributions to CC impacts confirms that the electricity grid remains the dominant source (Figure 5.9). In the smart charging configuration without BESS, grid electricity contributes 69 %, followed by wind (20 %) and PV installations (11 %). With BESS, the grid share declines to 65 %, while wind and PV installations contributions rise to 23 % and 11 %, respectively. BESS technologies (LFP and NMC) together contribute around 1 % or less. In the uni-directional with BESS configuration, PV installations and wind each contribute around 21 % and 20 %, with the grid reduced to 58 %. Without BESS, grid consumption increases to 62 %, with PV installations and wind each contributing 19 %. EV battery contributions remain negligible across all configurations. Benchmarking these results against the average Belgian grid mix (153 gCO2eq/kWh), all configurations yield substantial environmental improvements, reducing impacts by over 110 gCO2eq/kWh. These findings highlight the environmental benefits of the proposed energy system designs, even without BESS integration. 5.3.2 Uncertainty evaluation Global uncertainty assessment A Monte Carlo simulation with 10,000 iterations is conducted to assess the uncertainty of CC impacts across the four mezzo-level configurations. In this simulation, the allocation of EV battery production for stationary storage is simplified. Given its low contribution, this adjustment is not expected to affect results. Figure C.25 shows the normalized outcomes: smart charging with BESS achieves the lowest mean impact (35 gCO2eq/kWh), followed by smart charging without BESS and uni-directional with BESS (both 40 gCO2eq/kWh), and uni-directional without BESS (45 gCO2eq/kWh). These trends are consistent with the static LCA results (Section 5.3.1). However, none of the static results fall within the 95 % confidence intervals. One likely reason for this could be the use of a different electricity dataset in the simulation: the ecoinvent ‘market for electricity, low voltage’ (BE), which results in higher impacts than the modeled hourly mix. Figure 5.10: Comparison of normalized (per kWh) and unnormalized (total) distributions of climate change impact for four flexibility configurations in a mezzo-level electricity system in Belgium. Results are based on 10,000 Monte Carlo simulations. 5.3. MEZZO-LEVEL RESULTS 103 The narrow confidence intervals (e.g., [40, 40] or [34, 35]) suggest an artificially high level of certainty. One reason for this could be due to normalization per kilowatt-hour, which compresses output variability. Consequently, the confidence intervals presented in Figure C.25 do not reflect the full systemlevel uncertainty. To address this, unnormalized CC impact distributions are also evaluated (Figures 5.10 and C.26), which show greater variation. Standard deviations remain substantial (13-14 gCO2eq/kWh), and the overlap in means (e.g., 40 gCO2eq/kWh for both smart no BESS and uni-directional with BESS) confirms these configurations are statistically indistinct. A summary of descriptive statistics is presented in Table C.6. Notably, the Monte Carlo simulation produces some negative values, indicating net environmental benefits. While this may stem from inaccurate uncertainty data, further investigation is beyond the current scope. In summary, while the normalized Monte Carlo results support the static model findings, confirming that smart charging with BESS performs best, further analysis is required to assess smaller differences. To this end, a discernibility analysis is conducted to test whether observed differences are statistically meaningful. The results visualized in Figure C.28 are quantified in Table 5.4. Table 5.4: Quantification of discernibility results of the mezzo-level use case. Results are classified as similar when they lie within a range of ±5 gCO2eq/kWh compared to the baseline scenario. Baseline Compared Scenario Better than baseline (%) Similar (%) Worse than baseline (%) Uni – no BESS Uni – BESS 30.30 69.60 0.00 Smart – no BESS 87.60 12.40 0.00 Smart – BESS 99.90 0.00 0.00 Uni – BESS Uni – no BESS 0.00 69.60 30.30 Smart – no BESS 0.00 100.00 0.00 Smart – BESS 83.2 16.80 0.00 Smart – no BESS Uni – no BESS 0.00 12.40 87.60 Uni – BESS 0.00 100.00 0.00 Smart – BESS 27.70 72.20 0.00 Smart – BESS Uni – no BESS 0.00 0.00 99.90 Uni – BESS 0.90 16.80 83.20 Smart – no BESS 0.00 72.20 27.70 Based on the discernibility analysis presented in Figure C.28, the following conclusions can be drawn: 1. Uni - no BESS as baseline. Relative to Uni - no BESS,Smart - no BESS yields lower CC impacts in 87.60% of iterations and is similar in the remaining 12.40%; the baseline is never superior. Uni - BESS improves upon the baseline in 30.30% of iterations and is otherwise similar (69.60%), indicating a modest storage benefit under unidirectional charging. Smart - BESS is almost in all iterations preferable, with 99.90% of iterations resulting in lower CC impacts than the baseline and no cases worse or similar. 2. Uni - BESS as baseline. Against Uni - BESS,Uni - no BESS is worse in 30.30% of iterations and similar in 69.60%, reinforcing a moderate advantage for storage under unidirectional charging. The comparison with Smart - no BESS is indistinguishable across all iterations (100.00% similar), suggesting practical equivalence between these two configurations within the discernibility band. Smart - BESS outper- 104 CHAPTER 5. ASSESSING FLEXIBILITY SOLUTIONS IN ENERGY SYSTEMS forms Uni - BESS in 83.20% of iterations and is similar in 16.80%, with no cases worse, indicating a clear incremental benefit of combining smart operation with storage. 3. Smart - no BESS as baseline. When Smart - no BESS serves as the baseline, Uni - no BESS is worse in 87.60% of iterations and similar in 12.40%. Uni - BESS is indistinguishable in all iterations (100.00% similar), again pointing to near-equivalence between these two configurations under the chosen discernibility threshold. Relative to Smart - BESS, CC impacts are similar (72.20%), with 27.70% of iterations favoring Smart - BESS and no cases where it yields higher CC impacts than the baseline. 4. Smart - BESS as baseline. Compared to Smart - BESS,Uni - no BESS is almost always leading to higher CC impacts (99.90%) and never similar or better. Uni - BESS has higher CC impacts in 83.20% of iterations and similar in 16.80%, with only 0.90% showing a small advantage over the baseline. Against Smart - no BESS, CC impacts are largely similar (72.20%), with 27.70% of iterations leading to higher CC impacts and never lower, confirming a modest but consistent edge for storage under smart operation. Within the ±5gCO2eq/kWh discernibility band, the mezzo-level results confirm the qualitative ordering observed at the micro-level configurations combining smart charging with storage (Smart - BESS) provide the most robust reductions in CC impacts, followed by storage without smart operation (Uni - BESS) and then smart operation without storage (Smart - no BESS), with Uni - no BESS consistently highest CC impacts. However, the magnitude of differentiation is smaller at the mezzo-level: several pairwise contrasts are predominantly classified as similar (e.g., Uni - BESS vs. Smart - no BESS: 100% similar; Smart - BESS vs. Smart - no BESS: 72.20% similar), whereas the micro-level exhibited much stronger separation (e.g., Uni - BESS vs. Uni - no BESS was better in about 99.9% at micro-level but only 30.30% at mezzo-level). These findings indicate that while the rank order of configurations is stable across scales, scaling to the mezzo-level dampens contrast in CC impacts. One main reason for a less profound distinction of scenarios in the mezzo-level use cases could be the higher level of decarbonization and thus, an overall lower CC impact per system configuration. 6.1. USE CASE DESCRIPTION 111 6.1 Use case description 6.1.1 First life Even though the first life is excluded from the system boundaries, the usage of the first life influences the starting conditions of the SLB. To determine these conditions, the 2017 Renault Zoe ZE40 with a NMC811 is selected as it represents the Belgian EV fleet of 2019 [200]. Under the Worldwide Harmonized Light Vehicles Test Procedure (WLTP) cycles, the 2017 Renault Zoe ZE40 consumes 175 Wh/km [201]. According to national Belgian statistics, the lifetime mileage of vehicles is 133,668 km [23]. Consequently, the lifetime electricity consumption in the first life adds up to 23,392 kWh with a remaining capacity of 89 %. The battery manufacturing of the 41 kWh pack is assumed to take place in France. Containing 12 modules with cells manufactured in South Korea, the battery weighs 284 kg. Life cycle inventories for the battery components originate from Chordia et al. (2021), whereas the first usage is modeled using the cell datasheet of Lima (2021) [202,203]. As cathode active material, lithium nickel manganese cobalt oxide (LiNi0.8Co0.1Mn0.1O2, NMC811) is used, representing the expected, future, dominant battery chemistry in the global market and is modeled with primary data [79,204]. The cycle life of the battery in the first life is modeled following Lima (2021), which specifies the remaining battery pack capacity at the end of the first life to be 36.5 kWh [203]. 6.1.2 Use cases The residential use case (RES) represents a four-person Belgian household with a domestic 4 kWp PV installation in combination with SLB for stationary storage with an energy:power ratio of 2:1. The objective of the SLB is to maximize self-consumption of PV-generated electricity. As a result, 3 out of 12 battery pack modules of the 2017 battery are required. To set up the SLB in the household, a new battery management system and an electric cabinet are assumed. For the PLCA, the second life use stage impacts are modeled by converting the impacts of electricity production from the original ecoinvent dataset of 3 kWp PV installations of 30 years to twelve years, which represents the second-life assessed in the use cases [46,168]. Table 6.1: Main description of the residential use cases. (EV = Electric vehicles; SLB = Second-life batteries; RES = Residential use case; LTED = Lifetime delivered electricity; PV = Photovoltaic; n.a. = not applicable; adapted from [9,46,158,163,168]). Use case First life RES RES Battery EV SLB Benchmark Capacity (MWh) 0.041 0.0082 0.0082 Packs (numbers) 1 0.25 1 Lifetime (years) 9 12 15 LTED (MWh) 5.85 25.56 37.12 PV capacity (kWp) n.a. 4 4 PV electricity (MWh) n.a. 123.56 123.56 In the industrial use case (IND), behind-the-meter services, such as peak shaving or uninterrupted power supply, are provided at an industrial facility [102]. The used batteries, after removal, collection and dismantling, are mounted on racks in a container, which includes a cooling system. In a separate 112CHAPTER 6. UNLOCKING THE POTENTIAL OF SECOND-LIFE BATTERIES IN ENERGY SYSTEMS container, the required power electronics are hosted. In the PLCA, the same approach is followed as for the RES: the manufacturing impacts of the PV panels and the included system components are adjusted to match the second life. The last use case, the utility use case (UTI), shows SLB participating in front-of-the-meter service, e.g. interacting in the secondary reserve market. For this use case, the battery packs are mounted on racks in containers including a cooling system. Power electronics are located in a separate container [102]. The inclusion of power electronics is similar to the IND. In the UTI, the second life stage takes into account only the electricity losses due to charging and discharging. When participating in the Belgian secondary reserve market, a certain capacity of the batteries is contracted [205]. Afterwards, the distribution system operator charges and discharges the storage installations according to their needs. Due to technical performance and potential safety issues, the repurpose duration is 12 years in all use cases. The main underlying parameters are summarized in Table 6.1 and 6.2 and further details on second-life specific data are provided in the supporting material 1 of PAPER 4. Further details on assumed parameters, along with calculation results, are provided in the supporting material 1 and 2 of the same publication (PAPER 4). Table 6.2: Main description of the industrial and utility use cases. (EV = Electric vehicles; SLB = Second-life batteries; IND = Industrial use case; UTI = Utility use case; LTED = Lifetime delivered electricity; PV = Photovoltaic; n.a. = not applicable; adapted from [9,46,158,163,168]). Use case IND IND UTI UTI Battery SLB Benchmark SLB Benchmark Capacity (MWh) 1.24 1.24 20 20 Packs (numbers) 34 31 549 488 Lifetime (years) 12 15 12 15 LTED (MWh) 4,221.18 5,551.10 18,963.00 24,937.50 PV capacity (kWp) 1,000 1,000 n.a. n.a. PV electricity (MWh) 68,430 68,430 n.a. n.a. 6.2 Results Figure 6.1 illustrates the CC impact of SLB, broken down by life cycle stage and distributed over time. The battery pack is manufactured in 2023, followed by the repurposing stage, which includes collection, dismantling, and the repurposing process itself, in 2032. Operation and maintenance (O&M) take place between 2032 and 2044, after which the SLB undergoes EoL treatment in 2044. The rows of Figure 6.1 represent the three use cases modeled with the BAU scenario. In addition to results obtained using PLCA, the Figure includes a blue line plot indicating the CC impacts of SLB calculated using the ecoinvent 3.9.1 cutoff database. Applying different background scenarios leads to similar conclusions than those presented in Figure 6.1. Visualization of the results obtained using the ELEC and MOL scenarios are provided in Figure C.30 and C.31 in the Appendix C. 6.2. RESULTS 113 Figure 6.1: Climate change impact of SLB across different use cases (RES, IND and UTI) and based on the BAU scenario, presented over time. The midpoint LCIA method ’IPCC 2021, Climate change, GWP100a’ is used to compute the impacts. Results of other background scenarios and for the benchmark battery are provided in Appendix C. (RES = Residential use case; IND = Industrial use case; UTI = Utility use case; EoL = End-of-life; SLB = Second-life batteries). 114CHAPTER 6. UNLOCKING THE POTENTIAL OF SECOND-LIFE BATTERIES IN ENERGY SYSTEMS 6.2.1 Residential use case In the RES use case, the CC impact of the SLB under the BAU scenario amounts to 58.7 gCO2eq/kWh (Figure 6.1). Approximately 87 % of these impacts occur during the O&M phase, of which 98 % stem from electricity used for charging, generated by PV installations. Key contributors in the PV manufacturing include medium-voltage electricity from Asia (RAS region in REMIND) and solar-grade silicon. Manufacturing contributes 21 % to the total CC impact, with aluminium used for the battery housing and electricity from the Southeast Asia (OAS) region being primary drivers. The impacts from collection, dismantling, and repurposing, grouped as ‘repurposing’ in Figure 6.1, are minor, contributing less than 1 %. In contrast, EoL treatment offers a potential benefit, reducing total CC impacts by 6 % through the recycling of materials like aluminium, cobalt, nickel, copper, and lithium. The benchmark battery’s CC impact is 76.5 gCO2eq/kWh, higher than the impact of the SLB, mainly due to the full allocation of manufacturing emissions to the second-life application. Notably, the PV electricity used for charging contributes similarly for both battery types, 50.3 and 49.5 gCO2eq/kWh for the SLB and benchmark, respectively. Switching from the BAU to the ELEC or MOL background scenarios results in a reduction of 20–22 % in CC impacts. This reduction is primarily driven by a 23% decrease in the impacts associated with PV electricity generation, caused by changes in datasets such as ’market for electricity, medium voltage (region Asia in REMIND model)’ and ’market for silicon, solar grade’. Overall, the PLCA approach consistently yields lower overall impacts. When using the ecoinvent 3.9.1 database, the CC impact of the SLB in the RES use case rises to 91.2 gCO2eq/kWh. 6.2.2 Industrial use case For the IND under the BAU, the total CC impact is 75.2 gCO2eq/kWh. Of this, 68 % is caused by the PV-generated electricity, 14 % by the battery manufacturing, and another 14 % by the power electronics container. Most of the container’s impact is due to heat produced from hard coal, which is used in manufacturing components like the DC/DC converter and printed wiring boards. The stages of collection, dismantling, repurposing, installation, and maintenance contribute less than 1 % to the total impact. Recycling in the IND only reduces the CC impact by around 1 %, due to avoided primary production of key metals. Compared to the RES, the CC impacts of the IND in the manufacturing and O&M stages are on a comparable level. The additional battery housing in the IND contributes to the increase of CC impacts compared to the RES. Interestingly, the benchmark battery performs better than the SLB in this case, with a CC impact of 71.2 gCO2eq/kWh. This is due to the higher SoH of the benchmark battery, which requires fewer batteries to meet the same storage capacity. Furthermore, the reference flow (mass per functional unit) is greater for the IND than the RES, explaining the higher absolute emissions in the IND scenario. As in the RES, switching from BAU to MOL leads to a 20 % reduction in CC impacts, while switching to ELEC results in an 18 % reduction. The majority (80–86 %) of these reductions are due to lower impacts from PV-generated electricity. As shown in Figure 6.1, C.30 and C.31, all three background scenarios (BAU, ELEC, MOL) result in lower CC impacts compared to modeling with the ecoinvent database version. When using the ecoinvent 3.9.1 cutoff database, the CC impact of SLB in the IND use case increases to 112.4 gCO2eq/kWh. 6.2.3 Utility use case In the UTI, the SLB results in a CC impact of 78.5 gCO2eq/kWh. The majority of this impact stems from the power electronics container (39 %) and pack manufacturing (34 %) (see Figure 6.2). In contrast to the RES and IND, where the use stage with the PV-generated electricity for charging dominates all life cycle stages (accounting for 86 % and 68 % respectively), the electricity required to cover charging losses in the UTI accounts for only 16 % of the total CC impact. This difference is largely due to the 6.2. RESULTS 115 use of Belgian electricity in the UTI scenario. Within the Belgian electricity mix, the main contributors to CC impacts are natural gas used in combined cycle power plants and the treatment of blast furnace gas in electricity production. Similar to the RES and IND use cases, the contributions from collection, dismantling, repurposing, and maintenance are negligible—each contributing less than 1 % to the overall CC impact. The benchmark battery in the UTI scenario shows a lower CC impact of 53.2 gCO2eq/kWh. This is partly because benchmark batteries are lighter and deliver more energy over their full lifetime compared to SLBs in second-life applications, which experience higher energy losses during operation. Switching from the BAU to the ELEC and MOL scenarios results in a 14 % and 16 % reduction in CC impacts, respectively. A contributing factor to this decline is the 13 % decrease in the CC impact of the power electronics container. Notably, the lower impact of the dataset for the ‘printed wiring board, surfacemounted, unspecified, Pb-free’ is a key driver of this reduction. Among the three use cases, the difference in CC impact between the PLCA results and the ecoinvent 3.9.1 cutoff database is smallest in the UTI. When modeled with ecoinvent, the SLB in the UTI reaches a CC impact of 90.8 gCO2eq/kWh, compared to 78.5 gCO2eq/kWh in the BAU scenario. 6.2.4 Effects of the Belgian Pathways on SLB Impacts Figure 6.2: Climate change impact comparison of second-life and benchmark batteries considering BAU, ELEC and MOL paths. Underlying data are provided in Table S5 of S1 (CC = Climate change; RES = Residential use case; IND = Industrial use case; UTI = Utility use case; others = Collection, dismantling, repurposing, installation, and maintenance). 116CHAPTER 6. UNLOCKING THE POTENTIAL OF SECOND-LIFE BATTERIES IN ENERGY SYSTEMS Overall, changing from BAU to ELEC or MOL does not have a major impact on CC. Consequently, it highlights the importance of modeling the foreground system, whereas modifying the background is less relevant for CC of SLB. Additionally, it confirms the findings that the SLB is the preferable solution over benchmark batteries in the RES. However, in the IND and UTI the benchmark batteries emit lower CC impacts, regardless of future evolution of the background system. At the same time, it points out the importance of updated foreground datasets, e.g. for PV installations. A second observation is the similarity of the ELEC and MOL paths: as Figure 6.2 shows, the CC impact of the SLB and benchmark batteries are very similar for the ELEC and MOL paths. This is expected, as both scenarios aim to limit CO2emissions to the same level. Since this study focuses only on CC impact, the ELEC and MOL scenarios do not result in great changes. In contrast, the CC impact is higher in the BAU for two main reasons: first, the CC impact of the PATHS2050 BAU scenario is created to release more CO2emissions than the ELEC and MOL. Second, the BAU in this study is modeled with the NPi, which limits the GMST to 3.3 degrees by 2100 compared to the pre-industrial era, whereas the PkBudg1150 used to model the ELEC and MOL limits the GMST to between 1.6 and 1.8 degrees by 2100. Thus, it is unsurprising, that the CC impact of all batteries in the BAU are higher than impacts of batteries in the ELEC and MOL. Additionally, the only variation between the ELEC and MOL in this study is the different electricity mixes for Belgium. According to ecoinvent 3.9.1, the CC impact of the Belgian electricity mix is 167.9 gCO2eq/kWh [46]. By 2047, the CC impacts of the Belgian electricity mix in the BAU, ELEC and MOL paths drop to 39.9, 34.7 and 34.7 gCO2eq/kWh respectively (see Figure 6.3). Even though this represents a reduction of 75 % compared to the current mix, the difference between ELEC and MOL in 2047 is not very profound. Figure 6.3: Climate change impact comparison of one kilowatt hour electricity, low voltage BE for the BAU, ELEC and MOL paths resulting from the user-defined PLCA database. Moreover, the Belgian electricity mix is only used in the second life value chain in Belgium, whereas the main contribution activities of the whole battery supply chain, such as cell manufacturing or pack assembly remain outside of Belgium and thus, unaffected by the adjusted Belgian energy paths. To summarize, the sensitivity of the conducted PLCA towards the IAM scenarios appear to be much higher than towards the modeled Belgian energy pathways. 6.2. RESULTS 117 6.2.5 Other environmental impacts In addition to CC impacts, the environmental performance of SLB has been assessed across the 25 midpoint impact categories of the Environmental Footprint version 3.1. For each category, the use cases have been ranked from 1 (lowest environmental impact) to 3 (highest). Table 6.3 presents an overview of the ranking. A key observation is that the RES use case frequently outperforms the others, ranking lowest in 20 out of the 25 categories. Figure 6.4: Energy and material resource impacts and water use of SLBs across the three use cases (RES = Residential use case; IND = Industrial use case; UTI = Utility use case). 118CHAPTER 6. UNLOCKING THE POTENTIAL OF SECOND-LIFE BATTERIES IN ENERGY SYSTEMS Group 1: RES >IND >UTI: This pattern, where RES performs best, followed by IND and then UTI, is observed in 12 impact categories. Notably, this includes ADP_MM, as shown in the top chart of Figure 6.4. Compared to RES, the SLB impact in IND and UTI for ADP_MM is approximately 67 % and 145 % higher, respectively. A plausible explanation is that larger installations, as found in the UTI, require more materials and packaging, which is not offset by the higher electricity throughput. This suggests that smaller systems, such as those in RES, generally lead to lower environmental burdens. Other impact categories in this group include human toxicity (HTC and HTN), LU, AC, EUF and ETF (see Table 6.3). Group 2: RES >UTI >IND: In this second group, RES again shows the lowest impacts, but IND performs worst. This group includes eight categories such as CC, ERR, EUM and EUT, PM, and POF. In the very specific case of ERR, the differences are less pronounced than in the impact category presented in Group 1. For instance, in the case of ERR, IND and UTI exhibit impacts 11 % and 5 % higher than RES, respectively, ranging from 1.36 to 1.5 MJ/kWh (net calorific value). Group 3: UTI >RES >IND: This smaller group includes three categories, WU, OD, and ETF (organics), where UTI performs best. In terms of WU, the UTI system consumes only half as much water as the RES and IND systems. This translates into an increase in WU of approximately 117 % (RES) and 130 % (IND) compared to UTI (see Figure 6.4). Special cases: Two categories deviate from the patterns above. For CCB, UTI has the lowest impact, followed by IND and RES. For IR, the IND use case performs best, and UTI the worst. Table 6.3: Use case ranking by EF v3.1 midpoint impact categories (RES = Residential use case; IND = Industrial use case; UTI = Utility use case; CC = Climate change; CCB = Climate change: biogenic; CCF = Climate change: fossil; CCL = Climate change: land use and land use change; ERR = Energy resources: non-renewable; ETF = Ecotoxicity: freshwater; EUM = Eutrophication: marine; EUT = Eutrophication: terrestrial; IR = Ionising radiation: human health; OD = Ozone Depletion; PM = Particulate Matter Formation; POF = Photochemical Ozone Formation: Human Health; WU = Water use). First rank Second rank Third rank EF v3.1 Midpoint Impact Categories RES IND UTI AC, ETF, ETFI, EUF, HTC, HTCI, HTCO, HTN, HTNI, HTNO, LU, ADP_MM RES UTI IND CC, CCF, CCL, ERR, EUM, EUT, PM, POF UTI RES IND ETF, OD, WU UTI IND RES CCB IND RES UTI IR Overall, the CC impact performance of SLB varies substantially across use cases and is highly influenced by the life cycle stage and background scenario. In the RES, SLBs clearly outperform benchmark 6.3. DISCUSSION AND LIMITATIONS 119 batteries, primarily due to lower manufacturing burdens and the use of PV electricity, which, despite its contribution, benefits from decarbonization in future energy pathways. In contrast, in the IND and UTI applications, SLBs tend to underperform compared to benchmark batteries, largely due to higher energy losses, additional housing and power electronics infrastructure, and lower state-of-health requiring more material per functional unit. While background scenarios such as ELEC and MOL reduce overall impacts by up to 22 %, their influence is modest compared to the design and operation of the foreground system. Importantly, the similarity between ELEC and MOL highlights that domestic decarbonization of electricity alone does not shift the climate performance of SLBs, especially when critical upstream processes (e.g., battery manufacturing) remain tied to global supply chains. These findings emphasize the importance of system-level integration, use-case-specific design choices, and accurate modeling of operational parameters when evaluating second-life battery applications. 6.3 Discussion and limitations The selection of the battery chemistry for SLB and benchmark is important: For EV, the European Commission foresees NMC811 to be the dominating chemistry in 2040. For grid-scale and behind-the-meter storage, lithium-ion batteries with a share of at least 82% are expected to be the most dominate, with LFP chemistry being the most commonly installed cathode in 2040, followed by NMC811 chemistry [206]. Following those numbers, it would have been more representative to use an LFP instead of a NMC811 battery for stationary storage. However, the objective of this study is not comparing different battery chemistries, but to understand how the CC impact varies when comparing the benchmark with SLB. Additionally, the future dominant battery chemistry is already adopted in the study and thus, this chemistry can be a representation for future EVs. Different battery chemistries and sizes will result in different CC and other environmental impacts. Looking into other impact categories reveals different conclusions about which use case results in the lowest environmental impacts, where some relative changes are greater than others. Thus, this study also highlights the importance of not drawing conclusions based on CC impacts only, but also to take other impact categories into account to determine which SLB configuration results in the most environmentally friendly conclusion. Lowest CC impacts of the SLB are identified for the RES, followed by the IND and UTI. To also decrease the CC impact of the SLB in the IND and UTI, more environmentally friendly solutions for the power electronics container would be needed. Additionally, both assessments point out the importance of the use stage, as it accounts for between 12% and 86% of the PLCA. This emphasizes the importance of updated datasets in the use stage, and realistic performance parameters to calculate the delivered electricity. While JRC guidelines on use stage modeling for mobile batteries recommend including only electricity losses due to charging of the batteries, guidelines for stationary storage not applying a cut-off approach are less transparent [207,208]. In this PLCA, the use stages of the RES and IND take the total charged/discharged electricity into account, while in the UTI, only the losses associated with charging the batteries with the Belgian electricity mix are included. Furthermore, the reliability of this study could be advanced by an energy model or primary data, such as charging rates and numbers of cycles per use case, to obtain a more realistic use stage assessment, as exemplified by Terlouw et al. (2023) [209]. Along with a more realistic energy model comes the need to model different lifetimes for the use cases. The current model assumes a lifetime of 12 years for the SLB for each use case [99], while the benchmark is modeled with a lifetime of 15 years [21, 103]. Obtaining a realistic evaluation of use case-specific lifetimes would require further testing. A perturbation analysis investigating the sensitivity of the CC impact and LCOS towards lifetime is given in the supporting information 1 of PAPER 4. Regarding testing of the batteries after first life, the used electricity is considered lost in this study, although it could be recovered. Moreover, this study presents a 120CHAPTER 6. UNLOCKING THE POTENTIAL OF SECOND-LIFE BATTERIES IN ENERGY SYSTEMS PLCA of SLB in different use cases with changing background databases according to when the activities occur in time. According to Sohn et al. (2020), this study could be framed as dynamic scoping with partial dynamic life cycle inventories [210]. In order to obtain a comprehensive dynamic LCA, a combination of both the dynamic life cycle inventories with dynamic characterization factors, e.g. as demonstrated by Levasseur et al. (2010), remains open for investigation [211,212], which has been accomplished by Diepers, Müller, and Jakobs [213]. Moreover, this Chapter presents user-defined scenarios for the Belgian energy projections PATHS2050 to build prospective databases, including an updated electricity mix for Belgium in 2050, which is not available in the regular PREMISE databases. Providing open-access material to easily rebuild the userdefined scenarios will facilitate the utilization of this data also for other LCA practitioners. Additionally, the provided materials allow to easily modify and update the projections and thus the resulting databases. Thus, this research is not only a great addition to the evaluation of SLB, but also contributes to modeling the Belgian energy transition beyond the scope of this study. Last, the Chapter combines EMS and LCA using the premise package. While some parts of this work are very straightforward, leaving no room for interpretation, such as the matching of some processes from the EMS with LCA datasets, the selection of the IAM and its pairing with the respective foreground system are a subjective decision of the practitioner. All future projections of the PATHS2050 are modeled to reach climate neutrality by 2050. However, for the BAU scenario the IAM NPi is chosen, which can be identified as a mismatch. However, this pairing is done intentionally to cover one projection that is not based on a very optimistic modeling. Additionally, in combination with the other EMS, matched with an IAM depicting a rather optimistic future, the work covers a broader range of potential pathways. Additionally, the difference between the three selected scenarios, both on the SLB themselves and on the Belgian electricity mix, is thoroughly assessed and presented throughout this work. 7.2. ENVIRONMENTAL RESULTS 127 A second set of impact categories, decreasing as well in 2050 compared to 2014, show a slightly different tendency: While still below the impacts of 2014, some impact categories show lower impacts in 2030 compared to 2050. For example, water consumption declines from 2.9 m3in 2014 by around 50 % in 2030, while it is limited to about a 40 % decline by 2050. Reasons for this are either increased water consumption due to more grid consumption, e.g. in the low_flex, medium_flex and high_flex scenario or due to larger water consumption because more PV electricity consumption in the high_flex_PPH scenario. Figure 7.3 reveals similar observations for terrestrial acidification, freshwater eutrophication and human carcinogenic toxicity. 7.2.2 Negative implications on the environment The third observed tendency is, contrary to the CC impacts, LU, water consumption, etc., is that towards 2050, certain impact categories show an incline. Following Figure 7.4, this Section elaborates those tendencies, demonstrated with ADP_MM and ETT, but hold also true for ETM and ETF. Figure 7.4: Evolution of selected impact categories increasing over time for different flexibility scenarios in Belgium (PPH = High prosumer potential; COM = Commercial sector; RSD = Residential sector). The total ADP_MM increase from 1.1 in 2014 up to 1.5 gCueq/kWh in the high_flex_PPH scenario in 2050, representing an increase of more than 45 %. Between 60 and 85 % of total ADP_MM per scenario comes from the Belgian electricity grid mix, except for the high_flex_PPH scenario in 2050. The major contributors to ADP_MM in the Belgian electricity grid mix are nuclear power plants, onshore and offshore wind turbines, PV installations and the distribution network. More PV generated electricity and greater charging/discharging of the batteries translate into higher ADP_MM. Consequently, the scenarios 128 CHAPTER 7. DECENTRALIZED ROLL-OUT OF PHOTOVOLTAIC SYSTEMS with higher capacities of PV and battery installations result in greater ADP_MM. As per technology, at least 70 % of ADP_MM is generated by PV installations, even though compared on a kWh-basis, batteries correspond to more ADP_MM. Impacts on ADP_MM occur mainly in the commercial sector, which becomes particularly evident in the high_flex_PPH scenario in 2050, where the Belgian electricity grid mix contributes only 37 % of the total ADP_MM. At the same time, the ADP_MM of the infrastructure in the 2050 high_flex_PPH scenario doubles. This increase in ADP_MM can be linked to the increase of PV-generated electricity, corresponding to about 55 % of the total ADP_MM in the high_flex_PPH scenario. Also for TEX, over 75 % of the impacts in 2014 occurred due to the Belgian electricity grid mix, while all infrastructure impact is emitted by the PV installations. The ETT impact of the Belgian electricity grid mix in 2014 is dominated by combined heat and power cogeneration based on wood chips, nuclear power plants, hard coal, and lignite power plant imports from the Netherlands and Germany. Additionally, the transmission and distribution network contributes to ETT. Comparing the low_flex scenario in 2014 with the high_flex_PPH scenario in 2050, an increase of 66 % of ETT is observed. Similar observations can be made for the other scenarios: an increase in ETT over the years is also observed for the other scenarios (see Figure 7.4). The Belgian electricity grid mix is the main contributor in all scenarios, with at least 49 % of the total ETT. However, due to the nuclear and coal power plants phaseout in Belgium, the ETT of the future Belgian electricity grid mix is dominated by wind power plants and the transmission and distribution network. An exception to the contribution of the Belgian electricity grid mix is the 2050 high_flex_PPH scenario: here, the infrastructure alone of the high_flex_PPH scenario accounts for more than 70 % of total ETT, where over 90 % can be allocated to PV installations. This is due to the higher consumption of PV-generated and stored electricity. Except for 2014, 60% of the total ETT is caused by the commercial infrastructure. Additionally, the PV installations make up at least 75 % of ETT of the infrastructure of all years. The inverter and the electronic installations, followed by the single-Si wafer, account for most of ETT. Within the PV panel production for the wafer, copper, aluminum, steel, liquid argon, electronic and solar-grade silicon and the consumed electricity at medium voltage prevail in the ETT. Similar observations for ETT, but in different magnitudes, can be found for freshwater and marine ecotoxicity. 7.3. SENSITIVITY 129 7.3 Sensitivity Figure 7.5: Sensitivity factors for the high_flex_PPH scenario in 2050. Each investigated parameter is indicated twice: first, with a 10% reduction and then with a 10% increase (SEN_1&2: Energy density; SEN_3&4: Battery capacity (RES); SEN_5&6: Battery capacity (COM); SEN_7&8: Share of SLB; SEN_9&10: PV-generated electricity (RES); SEN_11&12: PV-generated electricity (COM); SEN_13&14: Solar irradiation; SEN_15&16: PV installation size). All selected sensitivity parameters pertain to the design and performance assumptions of renewable energy systems, with a particular focus on RET installations. The high_flex_PPH scenario for 2050, characterized by the highest level of RET integration, is used to demonstrate the model’s sensitivity. Sensitivity results for all other scenarios and 18 environmental impact categories are available in Supporting Material 1 and 2 of PAPER 5. Sensitivity parameters 1 to 8 affect battery installations, while parameters 9 to 14 target PV installations. Following Figure 7.5, the SFs for parameters 1 to 8 demonstrate a consistent trend: a 10 % reduction in battery-related parameters increases environmental impacts by approximately 10–15 %, while a 10 % increase yields a smaller, yet still notable, decrease of 5–10 %. This pattern is consistently observed across most impact categories, including CC impacts, IR, ETF, and ETM, suggesting that the model responds symmetrically but less strongly to parameter increases. For parameters 9 and 10, which vary PV electricity generation in the residential sector, a similar sensitivity range is observed. The relative importance of these parameters aligns with the fact that PV generation in the residential sector constitutes a smaller share of total generation compared to the commercial sector. Nevertheless, sensitivities remain within the 10–15 % increase and 5–10 % decrease range across all categories, indicating proportional model responsiveness. In contrast, sensitivities 11 and 12, representing changes in PV generation in the commercial sector, reveal more differentiated behavior. In SEN_11 (-10 % PV generation), the increase in SFs varies notably across impact categories: IR shows the highest sensitivity with a +15 % increase, while categories such as CC impacts, OD, EUM, ETF, ETM, HTC, LU, and WU experience increases between 10% and 15 %. Other categories, including OFH, PM, OFT, AC, FEW, ETT, HTN, ADP_MM, and MRF, register smaller increases, typically below 5 %. This spread underlines that a reduction in commercial PV 130 CHAPTER 7. DECENTRALIZED ROLL-OUT OF PHOTOVOLTAIC SYSTEMS installation output does not impact all categories equally, but instead amplifies specific midpoints more strongly, especially those related to ionizing radiation and resource use. Conversely, in SEN_12 (+10 % PV generation in the commercial sector), most categories exhibit a reduction in SF ranging from 7-9 %, with several categories showing even higher decreases. Specifically, IR, LU, and OD demonstrate the strongest reductions, with SFs declining by approximately 11-12 %, indicating a high sensitivity to increased commercial PV installation generation. Categories such as CC impacts, EUM, ETF, ETM, HTC, and WU also respond notably, with reductions generally between 8-9 %. Slightly smaller decreases (between 6-8 %) are observed in categories such as OFH, PM, AC, HTN, and ADP_MM. Sensitivities 13 and 14, which adjust solar irradiation, induces the most pronounced responses. A reduction of solar irradiation by 10 % results in a sharp SF increase exceeding 15 % for nearly all categories, excluding OD, IR, and LU, where lower increases are observed (between 10-15 %). Inversely, a 10 % increase in irradiation reduces SFs by 10-15 %, reinforcing the importance of this external, geography dependent parameter. The exception remains IR, which exhibits a more subdued response, further underscoring its distinct behavior. Sensitivity parameters 15 and 16 (PV installation size) also show substantial influence. A 10 % reduction increases SFs by more than 15 % for categories, apart from SOD, IOR, and LUS, while increases in installation size lead to moderate decreases (typically 10-15 %), except for IOR. To summarize, the model is shown to respond similarly to changes in the investigated parameters, except for the solar irradiation, which influences the model more than others. However, the source of this parameter enjoys a great reputation and represents the geographic conditions of Belgium [163]. 7.4 Discussion and limitations This study is subject to the methodological choices of the researchers and other limitations. Regarding methodological issues, different approaches for modeling SLB exist (Schulz et al., 2020) [214]. The scientific community has not yet reached a consensus about which modeling approach to follow for SLB. First, SLB can be integrated without any environmental burden as they were manufactured for use in their first life [102]. Another modeling approach suggests allocating all environmental impacts of the battery between the different applications. This could be done based on the remaining capacity at the end of each application or the delivered electricity [98]. Third, the allocation could be avoided by expanding the system [80]. For this study, the first approach is followed. By 2030 and 2050, batteries from 33,620 and 2,747,134 EVs of the Belgian EV fleet will reach their EoL [179]. As these batteries will become available for repurposing anyway, only impacts linked to SLB repurposing are included in this study. Another methodological aspect is the selection of the appropriate LCA method. For this study, an ALCA is conducted. As the study focuses on identifying the environmental impacts of decentralized PV installations combined with stationary battery storage and not to assess the changes in the energy markets due to the installation of decentralized energy systems, an ALCA is conducted. However, this study could have been further extended to include the consequences of introducing decentralized PV installations supported by stationary battery storage, hence assessing the environmental impacts of the changed Belgian electricity grid mix. In this case, a CLCA should be performed. Additionally, the high penetration of PV installations with battery storage for an isolated electricity system might not necessarily result in a more favorable solution than the existing electricity grid. Mathiesen et al. (2015) pointed out that electricity storage technologies should not be considered the main solution for the technical integration of fluctuating RET [215]. Further, they credit battery electric storage for short-term integration and grid stability but point out their high losses and costs of system integration compared to available alternatives. As a solution, Mathiesen et al. (2015) propose that coupling the electricity, heat, and transport sectors would allow the fuel-efficient inclusion of more RET [215]. To mitigate the environmental impacts of a large battery storage installation, an alternative could be to 7.4. DISCUSSION AND LIMITATIONS 131 consider the heating sectors when optimizing the electricity sector towards a system with higher PV installation penetration. Moreover, allowing higher flexibility by decentralized energy systems also affects the merit order of the Belgian grid electricity mix and changes the need for other energy generation capacities. This study is conducted from an attributional point of view; different quantities of PV and battery installations are assessed over time, focusing on the residential and commercial sectors. Contrary to that, the LCA can also be conducted from a macro-energetic point of view. Thereby, the focus is to understand the interaction of the decentralized technologies to allow different levels of flexibility and the Belgian energy system. Following the macro-energetic approach is expected to reveal the consequences of introducing decentralized technologies for the Belgian energy mix and lead to diverging results. Besides, special attention should be given when quantifying the materials required for PV installations and battery storage. In particular, silicon metal, indium, and gallium are used for various PV modules, while many batteries contain cobalt and natural graphite. All these substances are listed as critical raw materials by the European Union [216]. Besides the long transportation routes to the manufacturing location, these materials are sourced in countries exposed to several social risks. Therefore, the social impacts on the supply chain of raw materials should be further assessed, as well as the social impacts of rolling out decentralized technologies, such as the social acceptance of citizens [217,218]. In addition, the increased need for raw materials represent an opportunity to develop innovative businesses that recover materials from obsolete PV and battery installations. Finally, compared to the previous Chapters, the uncertainty of the background database is not evaluated. However, the assessment of various scenarios reflects different technological pathways projected towards 2050. While this Chapter may not feature a statistical analysis as detailed as in other Chapters, the results remain valid, especially with the inclusion of multiple 2050 scenarios capturing a broad spectrum of potential futures. 132 CHAPTER 7. DECENTRALIZED ROLL-OUT OF PHOTOVOLTAIC SYSTEMS Chapter 8 Conclusion, recommendations, and outlook Whether supported by Sustainable Development Goal 7 (“Affordable and Clean Energy”), the European Green Deal, REPowerEU, or the Fit for 55 package, the decarbonization of the energy system has become a core societal objective. One key pathway towards decarbonization is the increased deployment of renewable energy technologies such as wind and solar power. However, the integration of variable renewable energy sources requires additional measures to ensure security of supply. In this context, this thesis advances the application of LCA, a widely used method to quantify environmental impacts, to emerging energy technologies. Specifically, it focuses on wind and solar energy as generation sources, SLBs for stationary storage, and EVs providing flexibility services to support grid stability. As a case study, the analysis is contextualized with a focus on Belgium. The application of LCA to emerging energy technologies has been extensively explored in the scientific literature. In the case of wind energy, previous studies can be broadly categorized into three groups: (i) LCAs including spatial variability based on regional electricity production; (ii) analyses incorporating regional factors in LCAs such as transport distances and sea depth; and (iii) large-scale modeling approaches combining regionalized LCIs with fleet-level assessments. A key limitation of existing studies lies in the scalability of regional models beyond localized contexts. Regarding flexibility services provided by EVs, the literature remains scarce. Among the three identified LCA studies, only one, conducted in a German context, explored flexibility across different scales. For SLBs, more research exists, primarily examining residential and industrial applications. However, despite assuming SLB lifetimes of up to 12 years, none of the studies considered temporal dynamics in background systems. At a systems level, coupling LCA with ESMs is not new. Nonetheless, research is lacking on decentralized, country-level deployment of PV systems combined with rising EV shares and BESS. Furthermore, many studies maintain narrow scopes, often focusing solely on climate change impacts while neglecting other environmental categories. Based on the literature, four research gaps were identified, leading to the formulation of three key challenges: (i) the need for temporally and geographically advanced life cycle inventories, (ii) the consideration of multiple system scales, and (iii) the demand for updated and flexible LCA modeling frameworks. This thesis addresses these challenges through the development of four novel LCA models. In a first step, LCI data are more temporally refined. In this context, the refinements can be distinguished into shortand long-term refined LCI. For the long-term refined LCI, the temporal evolution of SLB processes/stages is modeled using PLCA. Within this framework, primary LCI data for setting up an SLB value chain in Belgium is collected. For this purpose, user-defined scenarios based on the Belgian 133 134 CHAPTER 8. CONCLUSION, RECOMMENDATIONS, AND OUTLOOK PATHS2050 projections in PREMISE are built and made publicly available. Contrary to the long-term refined LCI, the short-term refined LCI data covers the modeling of hourly CC impacts of the Belgian electricity mix. With regard to the geographical refined LCI, a model was updated with further geospatial data, capable of assessing environmental impacts of onand offshore turbines all over Europe. Along with this model, the AEP of the existing EU wind fleet as well as the future BE offshore wind fleet is calculated. Another advancement of this thesis aims at overcoming the challenge of including differences in scales. To this end, the thesis presents a comparative LCA framework to assess flexibility services delivered at a micro-level and a mezzo-level use case. Including a statistical comparison allows evaluation of different charging strategies, such as uni-directional and smart charging. Taking one scale up, the developed wind LCA model is applied to the Belgian wind fleet and an analysis of its drivers and the behavior of fleet impacts is conducted. Next to wind energy, the thesis also covers the assessment of solar energy: as the last part of the thesis, the future impacts of decentralized, national roll-out of PV installations combined with EVs providing flexibility services and BESS are investigated. In this regard, the Belgian TIMES model is combined with LCA, where future market datasets for PV, SLB and BESS are built. 8.1 Contextualization This section reflects on the key findings, methodological advancements, and limitations of the research presented in this thesis. It puts the results within the broader context of Belgium’s evolving energy policy and climate ambitions, while also examining how emerging technologies such as offshore wind, SLB, and EV flexibility services may contribute to achieving decarbonization goals. The contextualization is structured around three main themes: (i) contextualizing the results in light of Belgian energy transition, (ii) assessing the methodological contributions to LCA, particularly through digital tools, geo-spatial and temporal refinement, and hybrid modeling approaches, and (iii) critically reflecting on the evolution of the research process, including trade-offs and choices made throughout the course of the PhD. 8.1.1 Implications for the Belgian energy transition Recent geopolitical developments have led to a reversal of Belgium’s initial plan to phase out nuclear energy by 2025. Instead, the updated nuclear policy includes the lifetime extension of the Doel 4 and Tihange 3 reactors until 2045, as well as the option to build new reactors [13]. This policy shift is intended to support the gradual reduction of Russian fossil gas imports. In parallel, Belgium reaffirmed its climate ambitions by committing, alongside countries such as the Netherlands, Luxembourg, Germany, France, Austria, and Switzerland, to achieve climate neutrality by 2035 [1]. To advance this goal, Belgium joined the North Seas Energy Cooperation, a collaboration with Denmark, France, Germany, Ireland, Luxembourg, the Netherlands, Norway, and the European Commission, aiming to accelerate offshore renewable energy development in the North Seas region [219]. This thesis assessed the CC impacts of a future Belgian offshore wind fleet, including an estimation of its potential electricity production. Results suggest that by 2050, the combined onshore and offshore wind fleet could generate up to 116.4 TWh of electricity annually. Given Belgium’s average gross final energy consumption over the past 25 years (393,101 GWh), this would represent approximately 30 % from wind energy alone. Currently, the renewable share stands at around 16%, including all renewable technologies. According to its NECP, Belgium plans to increase its installed wind capacity from 5.6 GW to 15.6 GW by 2040, including 7 GW from offshore wind [14,194,195]. However, some sources indicate an 8 GW target for offshore capacity, reflecting a data discrepancy. To reach the capacity projected in this thesis by 2050, an additional 5.3 GW of offshore installations will be needed. This long-term goal, spanning 25 years, roughly equivalent to the lifetime of a wind turbine, also implies that all turbines currently installed will have been decommissioned by then. This opens possibilities such as repowering, 8.1. CONTEXTUALIZATION 135 which lies beyond the scope of this study. Furthermore, the focus of this thesis is limited to assessing future offshore wind. Onshore potential is not considered, though ramping up onshore capacity could also contribute to achieving NECP targets and reducing CC impacts. Beyond wind energy, this thesis examines the role of decentralized PV systems supported by BESS and EVs providing flexibility services. In the most optimistic scenario (high_flex_PPH), electricity production from decentralized PV installations could reach 20.5 TWh by 2050. Using the same historical energy consumption reference, this corresponds to an additional 5 % renewable share. Placing these figures in the context of the Belgian NECP shows that increasing the wind and solar deployment alone could help meet the 2030 target of a 33 % renewable share. However, the deployment trajectories modeled in this thesis extend only to 2050. This reinforces a key message for policymakers: achieving decarbonization targets for 2040 and 2050 will require further renewable energy deployment beyond offshore wind and solar. As more electricity from renewable sources enters the grid, the need for energy storage and flexibility services will grow. One promising avenue is the roll-out of SLB. To reduce the CC impacts of SLBs to levels comparable with benchmark batteries, attention must be directed toward lowering emissions from power electronics, container, and pack manufacturing. Addressing this would support the broader integration of renewable electricity, as demonstrated in Chapter 6. Another important element is the integration of EVs as providers of grid flexibility. Chapter 5 shows that the CC impact of the allocated EV battery production is negligible. Furthermore, the high_flex_PPH scenario illustrates how widespread EV flexibility services can help reduce CC impacts (see Chapter 7). However, Chapter 2 highlights ongoing scientific debates over whether grid services affect EV battery health. Despite this uncertainty, promoting smart charging over uni-directional charging is environmentally preferable, as it enables greater accommodation of renewable electricity and reduces curtailments. Key challenges include determining how to incentivize EV users to participate in smart charging, especially when not charging at home, and establishing a robust business ecosystem for such services. Moreover, it is important to note that this discussion has focused primarily on CC impacts. However, as Chapter 7 demonstrates, reductions in CC impacts may come at the expense of other environmental impact categories, including ADP_MM, ETT, ETM, and ETF. Finally, this thesis outlines potential pathways to decarbonize the energy and transportation systems from a technical perspective. Regardless of technological feasibility, an essential first step in reducing environmental impacts lies on the demand side. Rather than prioritizing technical solutions that accommodate increasing energy demand, initial efforts should focus on limiting, or ideally reducing, overall energy consumption. Such reductions need not rely solely on behavioral changes or cuts in usage but can also be achieved through improvements in efficiency. While a detailed discussion of demand-side measures falls beyond the scope of this thesis, it is important to emphasize that effective decarbonization should align with the first “R” of the waste hierarchy: “Reduce.” Within this context, the size of current and future EV batteries warrants critical attention. Originally intended to address range anxiety, battery capacities in EVs have grown substantially, with more and more EV models approximating or exceeding a battery capacity of 90 kWh. For instance, the Mercedes-Benz EQE is equipped with a 90 kWh battery, enabling a range of over 600 km on a single charge [220]. While such extended ranges may be justified in specific use cases, it remains questionable whether the average EV user requires such large default battery capacities, especially given their considerable environmental impacts. 8.1.2 Improved LCA methodology Over the past decade, the LCA community has experienced a rapid digital transformation. Driven by a vibrant open-source movement, the field is shifting away from monolithic, commercial software solutions toward more flexible and customizable approaches. This transition has empowered researchers to process larger datasets, build more sophisticated models, and visualize results in novel ways. However, 136 CHAPTER 8. CONCLUSION, RECOMMENDATIONS, AND OUTLOOK this flexibility also introduces new challenges for LCA practitioners striving to maintain methodological consistency and transparency. This thesis benefited from the advances in open-source LCA tools, enabling more refined LCI data collection. One key contribution lies in the use of temporally resolved LCI data. Chapter 6 demonstrates the advantages of long-term temporal refinement, particularly Figures C.32 and 6.1, which highlight the benefits of modeling long-term deployment dynamics. In contrast, short-term temporal refinement shows more limited advantages. In systems with low renewable capacity and high grid dependency, such as the micro-level use case, benefits from smart charging and BESS are observable (Figure 5.8). However, in the mezzo-level case, characterized by higher renewable shares, differences between configurations are less pronounced (Figure 5.9). With respect to spatially resolved LCI data, Chapter 4 reveals that the influence of geo-specific components on overall CC impacts is relatively minor when compared to major contributors such as the rotor, tower, nacelle, and foundation. Still, when normalized per functional unit (e.g., per kWh of electricity), regional differences do emerge, primarily due to variations in AEP between turbines. Thus, AEP proves more influential on impact outcomes than spatial refinement of component inventories. A broader methodological concern raised in this thesis relates to the implications of selectively applying LCI refinements. If only certain technologies benefit from more detailed data, this may introduce bias in comparative assessments, particularly when these models are integrated into larger systems, such as ESMs. This highlights a need for harmonized improvements across technologies to avoid skewed conclusions. Another critical methodological issue concerns the comparability of assessments across different scales. A central question is whether different scales, such as microand mezzo-level systems, can be meaningfully analyzed within the same LCA study, or whether they necessitate separate assessments due to differing goals and scopes. Chapter 5 illustrates this issue clearly. Although the systems analyzed at micro and mezzo levels are structurally similar, sharing generation technologies, storage, and flexibility services, they are modeled in separate LCAs. While comparisons across configurations within a single level are valid, cross-scale comparisons (e.g., claiming that smart charging with BESS in the mezzo-level has lower CC impacts than in the micro-level) are not methodologically sound. This reflects a structural limit in the comparative power of LCA across scales. Contrary to a simple static LCA, this thesis outlines evaluations to assess, present, and communicate uncertainty data of stochastic LCAs. Due to the difference in the study design, the chosen uncertainty assessments vary per Chapter. Nevertheless, it does not hamper the reliability and validity of this thesis. An open question in LCA is how to determine whether differences in environmental impact results are environmentally significant. LCA outcomes are calculated by multiplying the inventory matrix with a characterization matrix, resulting in quantified midpoint impacts such as CC impacts in kilograms of CO2-equivalent. However, assessing whether a given reduction in CC impacts is meaningful in environmental terms remains challenging. While statistical tools can evaluate whether a difference is robust or uncertain, they do not indicate whether the change meaningfully contributes to CC mitigation. Midpoint indicators such as the Global Warming Potential over a 100-year timeframe (GWP100) express relative contributions to global warming but lack defined thresholds for environmental relevance. To evaluate such significance, one would ideally need to relate LCA results to absolute climate science, such as remaining carbon budgets or temperature-based targets from integrated climate models. However, such integrative approaches are not yet standard practice in LCA and are not supported by existing methodological frameworks such as the EF method or ISO 14040/44. Nonetheless, the use of GWP100 based on the IPCC 2021 model is currently considered scientifically robust and is recommended for midpoint-level assessment of climate change impacts [221]. This method is also implemented in EF version 3.1, which is applied throughout this thesis. For further guidance on the methodological robustness of this and other impact categories, the recommendations outlined in Hauschild et al. (2013) provide a sound foundation [221]. Thus, although it remains difficult to determine