Modeling and Optimizing the Logistic Routing of Biomass Side-Flows for Circular Bioeconomy : Applying a Genetic Algorithm-Based Discrete Event Simulation Approach

Kasper Eloranta MODELING AND OPTIMIZING THE LOGISTIC ROUTING OF BIOMASS SIDEFLOWS FOR CIRCULAR BIOECONOMY Applying a Genetic Algorithm-Based Discrete Event Simulation Approach Faculty of Management and Business Master of Science Thesis Examiners: Senior Research Fellow Ulla Saari and Professor Juho Kanniainen September 2023 ABSTRACT Kasper Eloranta: Modeling and Optimizing the Logistic Routing of Biomass Side-Flows for Circular Bioeconomy: Applying a Genetic Algorithm-Based Discrete Event Simulation Approach Master of Science Thesis Tampere University Master’s Programme in Industrial Engineering and Management September 2023 The acceleration of climate change has led regulators to impose constraints and incentives, encouraging both companies and individuals to align their actions with the goals of a green transition. There's a growing imperative to reduce waste and emissions while simultaneously maximizing the utilization of waste and by-products as valuable resources. These trends are driving societies toward adopting principles of the Circular Economy (CE). As the consumption of fossil energy sources needs to be reduced, the development and production of renewable energy sources become essential. Biogas, a form of renewable energy, offers an efficient way to utilize agricultural bio-based waste and side-flows, contributing to waste reduction and providing an alternative to fossil fuels. This approach aligns with the principles of the Circular Bioeconomy (CBE). However, establishing a circular supply chain for biogas production from biomasses is a complex endeavor. To ensure effective and sustainable operations within this system, logistics planning is crucial. This involves determining where, when, and how much biomass should be imported and transported for biogas production. It presents a multi-objective optimization problem, aiming to guarantee continuous biogas production while minimizing the costs associated with transporting biomass flows and other occurred inefficiencies. This study models and simulates a circular supply chain that produces biogas from agricultural biowastes and side-flows. The supply chain in question encompasses a biogas plant, which is currently in the design phase but will later be constructed in the Kanta-Häme region of Finland. It also involves farms located in nearby areas, which are potential sources of biomass. The modeling aspect aims to optimize the logistics of collecting and transporting various types of biomasses from the farms to the biogas plant for use in the biogas production process. To achieve this optimization, a simulation model that utilizes a genetic algorithm (GA) has been developed. The objectives of the study include generating knowledge about both general and case-specific factors that should be considered when planning logistic routing to align with the principles of the Circular Bioeconomy (CBE). Another objective is to create an optimized routing plan for collecting resources for biogas production at the plant. Additionally, this study seeks to contribute to understanding of how time-critical factors, such as the quality of biomasses in terms of their energy potential, impact the optimization of logistic routing. The study successfully maps the biomass potentials near the biogas plant and develops a methodology for optimizing the logistic routing of biomasses using the simulation model and the genetic algorithm. The study achieves the objective of generating an optimized routing proposal for the biogas plant in the case study, ensuring a continuous biogas production process. However, it is important to note that the results of the study may be suboptimal due to the complexity of the optimization problem and the time constraints of the research. The time-critical nature of biomasses should be considered when planning logistic systems for biogas production to avoid unnecessary costs. Limitations of the study include the use of non-farm-specific datasets, simplifying assumptions made in the simulation model, and time constraints that influenced the optimization stopping criteria and the potential for suboptimal results. Further research on this topic is necessary, either with more advanced methods or by allocating more computational time for optimization. Keywords: Circular bioeconomy, logistic routing, optimization, biogas production, circular supply chain, simulation, genetic algorithm The originality of this thesis has been checked using the Turnitin OriginalityCheck service. TIIVISTELMÄ Kasper Eloranta: Biomassasivuvirtojen logistisen reitityksen mallinnus ja optimointi biokiertotalouden kontekstissa: geneettistä algoritmia käyttävän diskreetteihin tapahtumiin pohjautuvan simulaatiomenetelmän hyödyntäminen Diplomityö Tampereen yliopisto Tuotantotalouden DI-ohjelma Syyskuu 2023 Ilmastonmuutoksen kiihtyminen on johtanut lainsäätäjät asettamaan rajoituksia ja kannustimia, jotka rohkaisevat yrityksiä ja yksilöitä mukauttamaan toimintaansa vihreän siirtymän tavoitteiden mukaisesti. Jatkuva kasvava tarve vähentää syntyviä jätteitä ja päästöjä samanaikaisesti jätteiden uudelleenkäytön ja hyödyntämisen maksimoiden vie yhteiskuntia kohti kiertotaloutta (engl. circular economy). Fossiilisten energialähteiden vähentämisen tarve synnyttää tarvetta kehittää ja tuottaa uudenlaisia uusiutuvan energian lähteitä. Biokaasu on tällainen uusiutuvan energian muoto, mikä mahdollistaa tehokkaan tavan maatalousjätteiden ja -sivutuotteiden hyödyntämiseen samalla vähentäen syntyvien jätteiden määrää ja tarjoten kestävän vaihtoehdon fossiilisille polttoaineille lähestyen biokiertotaloutta (engl. circular bioeconomy). Koska biokaasua tuottavat toimitusketjut ovat luonteeltaan monimutkaisia, niiden logistisia toimintoja on optimoitava jatkuvuuden turvaamiseksi. Suunnittelu kattaa päätökset siitä, mistä, milloin ja miten paljon eri tyyppisiä biomassoja tulisi noutaa ja kuljettaa biokaasulaitokselle kulutettavaksi biokaasun tuotannossa. Tällainen toimitusketjun logistiikka muodostaa monitavoitteellisen optimointiongelman, jonka tavoitteena on biokaasun tuotantoprosessin jatkuvuuden turvaaminen samalla minimoiden logistisen kuljetuksen sekä muiden mahdollisten tehottomuuksien aiheuttamat kustannukset. Tässä tutkimuksessa mallinnetaan ja simuloidaan toimitusketjua, joka tuottaa biokaasua maataloudessa syntyvistä biojätteistä sekä sivuvirroista. Tarkasteltava toimitusketju sisältää suunnitteluvaiheessa olevan biokaasulaitoksen, joka tullaan sijoittamaan Kanta-Hämeen alueelle, sekä lähialueella biomassoja tuottavat tilat. Mallinnuksen tavoitteena on optimoida toimitusketjun logistinen reititys eri tyyppisten biomassojen keruun ja kuljetuksen suhteen. Optimoinnissa hyödynnetään geneettistä algoritmia simulaatiomallin kustannusfunktion minimoinnissa. Tutkimuksen tavoitteet kattavat ymmärryksen tuottamisen niistä yleisistä ja tapauskohtaisista tekijöistä, jotka on huomioitava toimitusketjun logistiikkaa optimoidessa varmistaakseen toimintojen yhteensopivuuden biokiertotalouden kanssa. Tavoitteena on myös tuottaa optimoitu reitityssuunnitelma kohdeyritykselle, joka vastaa biokaasulaitoksen suunnittelusta ja rakennuttamisesta. Reitityssuunnitelman tavoitteena on tuottaa kohdeyritykselle tietoa siitä, miten, mistä ja miten paljon eri biomassoja tulisi hakea. Lisäksi tutkimus tavoittelee ymmärryksen lisäämistä biomassojen sisältämien aikakriittisten tekijöiden vaikutuksesta logistisen reitityksen optimointiin. Tutkimus onnistuu kartoittamaan laitoksen lähialueen biomassat sekä tuottamaan metodologian biomassojen logistisen reitityksen optimointiin osana biokaasun tuotantoa hyödyntämällä kehitettyä simulaatiomallia ja geneettistä algoritmia. Tutkimus saavuttaa tavoitteen optimoidun reitityksen tuottamisesta, mikä turvaa biokaasun tuotantoprosessin jatkuvuuden. Optimoitu reititys ei kuitenkaan välttämättä saavuta globaalia optimia. Biomassojen aikakriittinen luonne tulisi ottaa huomioon biokaasua tuottavien toimitusketjujen logistiikkaa suunniteltaessa välttääkseen ylimääräisiä kustannuksia. Tutkimuksen rajoitteisiin kuuluvat lähtödatan tilakohdattomuus, mallinnuksessa tehdyt välttämättömät yksinkertaistavat oletukset sekä tutkimusajan rajallisuus optimointiongelman monimutkaisuuteen nähden. Jatkotutkimusta aiheesta tarvitaan, joko käyttämällä edistyksellisempiä ja tehokkaampia menetelmiä tai allokoimalla tutkimukselle enemmän aikaa, jotta tulosten optimaalisuus voidaan varmistaa pidemmällä laskenta-ajalla. Avainsanat: Biokiertotalous, logistinen reititys, optimointi, biokaasun tuotanto, kiertotalouden toimitusketju, simulointi, geneettinen algoritmi Tämän julkaisun alkuperäisyys on tarkastettu Turnitin OriginalityCheck –ohjelmalla. PREFACE These last five years studying at Tampere University have been an incredible learning journey for me. I have developed my skills in solving surprisingly complicated problems by trusting my rational and logical thought process, which has developed to an extent of which I am proud. More importantly, I have evolved as a person, friend, and companion. Every single person who has had even the slightest impact on my journey deserves acknowledgment of my appreciation and thankfulness. The study I am happily wrapping up today, at some level, summarizes the key driving interests that motivate me to pursue greatness. It includes elements representing the fundamental drive for me to navigate the complicated world through logic and numerical analysis, which was one of the main reasons I applied for and studied the first year of my studies in a program of technical physics and mathematics. During that first year, I realized I wanted to utilize my skills and interests in a field that enables me to provide more practical impacts and benefits to the companies or organizations I work for or with. That is why I decided to apply to the program of Industrial Engineering and Management, and now, four and a half years later, I am writing the preface of my master’s thesis, after which I will graduate from that program. My enthusiasm for data science and the utilization of advanced computational methods to provide real-world business value is also evident within this study. If I am lucky, someday I might get to work with these topics I feel passionate about. We shall wait and see what the future holds for me. I want to thank my supervisors at HAMK Smart, Olli Koskela and Iivari Kunttu, for providing me with this opportunity to conduct my study within an interesting project. Supervisors from Tampere University, Ulla Saari and Juho Kanniainen, also deserve great thanks for providing guiding principles and significant assistance to me in the process of writing this thesis. I am grateful for the friends I have made and have managed to stay in contact with during these busy years of studying. Thanks for providing opportunities to relax from studies for the most interesting ways of spending time. Stay just the way you are, as awesome. Lastly, I want to thank my beloved fiancée, Eeva, for supporting me and always counting on me during these years. Your unceasing support for my ambitious endeavors is admirable. May we continue this tremendous and exciting journey of ours with the excitement and trust for the future. In Tampere, Finland, 20.9.2023 Kasper Eloranta CONTENTS 1 INTRODUCTION .................................................................................................. 1 1.1 Research background .......................................................................... 1 1.2 Scope and objectives of the research .................................................. 3 1.3 Research problem and questions ......................................................... 5 1.4 Structure of the study ........................................................................... 5 2 CIRCULAR BIOECONOMY .................................................................................. 7 2.1 From bioeconomy and circular economy to circular bioeconomy.......... 7 2.2 Conceptual framework for circular bioeconomy .................................. 10 2.3 Circular business models for valorizing agricultural biowaste and sideflows in biogas production.................................................................................. 16 2.3.1 Anaerobic digestion .................................................................... 17 2.3.2 Digestates as fertilizers ............................................................... 21 2.4 Supply chain of the biogas production ................................................ 23 2.5 Success and risk factors of the biogas supply chain........................... 29 2.6 Synthesis of literature review ............................................................. 30 3 LOGISTIC ROUTING OPTIMIZATION OF BIOMASS SIDE-FLOWS – METHODOLOGY ....................................................................................................... 33 3.1 Research design and strategy ............................................................ 33 3.2 The problem of biomass routing and biogas production .................... 36 3.3 Data collection and description .......................................................... 40 3.4 Data procession ................................................................................. 41 3.5 Model development ............................................................................ 50 3.5.1 Simulation approach to biomass routing optimization .................. 52 3.5.2 Genetic algorithm in optimization ................................................ 57 3.6 Validity and reliability of the methodology ........................................... 61 4 RESULTS ........................................................................................................... 63 4.1 Optimized routing ............................................................................... 63 4.2 Optimized routing under the assumption of biomass time-criticality .... 67 4.3 Benchmarking of results ..................................................................... 71 4.3.1 Benchmarks of the optimized routing .......................................... 71 4.3.2 Benchmarks of the optimized routing under the assumption of biomass time-criticality .............................................................................. 73 5 DISCUSSION...................................................................................................... 75 5.1 Comparison of results ........................................................................ 75 5.2 Key findings ....................................................................................... 77 6 CONCLUSIONS .................................................................................................. 80 6.1 Theoretical and practical implications ................................................. 80 6.2 Quality assessment and limitations of the study ................................. 81 6.3 Ideas for future research .................................................................... 82 REFERENCES....................................................................................................... 84 PICTURE LIST Figure 1. The research focus of the study. .................................................................... 4 Figure 2. Conceptual model for the circular economy (CE). Graph adapted from Stahel (2016) and Korhonen et al. (2018). ........................................... 14 Figure 3. Conceptual model for the circular bioeconomy (CBE). Graph adapted from Wei et al. (2022). .......................................................................... 16 Figure 4. Example illustration of the case company’s circular supply chain. For circular bioeconomy supply chain activities, see Ponjavic et al. (2020). For more details about the selected customer, see Jumppanen (2023). .............................................................................. 24 Figure 5. Framework of the factors within the biogas supply chain that can be improved with developed methods. ...................................................... 31 Figure 6. An example illustration of a partial solution to the Biomass Routing and Biogas Production Problem (BRBPP). ................................................. 39 Figure 7. An example of the form of unprocessed data. .............................................. 41 Figure 8. An example of the data pre-processing phase. ............................................ 42 Figure 9. An example of the filtering phase in data processing. .................................. 43 Figure 10. An example of the Elbow method in validating the number of clusters. ...... 44 Figure 11. An example of the initial clustering phase in data processing. .................... 45 Figure 12. An example of the methodology of cluster thinning. ................................... 46 Figure 13. An example of the nearest clusters to fulfil resource needs. ....................... 48 Figure 14. An example of the nearest clusters plotted within a map. ........................... 48 Figure 15. Pick-up sites approximated from biomass datasets. ................................... 49 Figure 16. Routing optimization progress. ................................................................... 64 Figure 17. Routing optimization progress under the assumption of biomass timecriticality. .............................................................................................. 68 ABBREVIATIONS AND NOTATIONS AD Anaerobic Digestion BCP Biomass Collection Problem BRBPP Biomass Routing and Biogas Production Problem CE Circular Economy CBE Circular Bioeconomy CBM Circular Business Model CSR Corporate Social Responsibility EOQ Economic Order Quantity EPQ Economic Production Quantity GA Genetic Algorithm GHG Greenhouse Gases HAMK Häme University of Applied Sciences HRT Hydraulic Retention Time LuKe Natural Resources Institute Finland OLR Organic Loading Rate RL Reinforcement Learning TS Total Solids TSP Travelling Salesman Problem VRP Vehicle Routing Problem VRPTW Vehicle Routing Problem with Time Windows ∆𝑚𝑡,𝑖 an accumulation of biomass at the day 𝑡 on the pickup site 𝑖 𝑚𝑡 a daily expected accumulation of biomass at the site 𝑖 𝑀𝑖 the annual accumulation of grass or straw at site 𝑖 𝜀𝑖 randomized vector determining the days of biomass jumps at pickup sites producing grass and straw 𝑟𝑙,𝑑 biomass weekly volume loss coefficient due passive drying 𝑉𝑏,𝑤 the volume of biomass 𝑏 after 𝑤 weeks of drying 𝑟𝑚,𝑑 biomass weekly moisture decrease coefficient due passive drying 𝑀𝑏,𝑤 the level of moisture of biomass 𝑏 after 𝑤 weeks of drying ∆𝑡𝑖,𝑙 the pickup duration for the pickup site 𝑖 with a load level of 𝑙 𝑇 setting time for pickup operation 𝑣𝑚 collection rate of biomass 𝑚 𝑆𝑝 size of the population utilized within GA’s optimization 𝑁𝑔 number of genes within a solution proposal TC total costs of the system caused during a simulation period 𝑐𝑜𝑣𝑙𝑑 the cost of one day of overload at a pickup site OVLDi the number of overloading days at pickup site 𝑖 Cvehicles costs caused by vehicles during a simulation period 𝑝𝑓 the price of fuel 𝑐𝑓 the assumed fuel consumption of vehicles ODO𝑖 the total mileage driven by vehicle 𝑖 𝑛 the number of vehicles 𝑐𝑜𝑣𝑡ℎ the assumed cost of overtime work OVT𝑖 the total overtime work of vehicle 𝑖 𝑐𝑤𝑣 unit cost for a vehicle visiting wrong type of pickup site WVi the number of wrong visits carried out by vehicle 𝑖 Cbgp costs caused by emerged inefficiencies within the biogas production 𝑐𝑝𝑑𝑠𝑡 assumed unit cost of production stoppage due biomass shortage PDST the number of production stoppages occurred during the simulation c𝑜𝑣𝑓 the cost of one day of overfilling biogas plant’s storage OVF the number of days of overfilled storage within the biogas plant c𝑒𝑥𝑖𝑚 the cost of biomass import after annual demand is satisfied EXIM the number of unnecessary imports occurred within the simulation 6 The third chapter presents the research methodology for this study, including a conceptual description of the optimization problem of biomass routing and biogas production, data collection, description, processing, and model development. It covers topics such as processing and visualizing data in Python (Magiya, 2019), utilizing the KMeans clustering algorithm (Jeffares, 2019), the SimPy library for discrete-event simulation (Team SimPy, 2020), and the genetic algorithm used in optimization (Gracia et al. 2014; Katoch et al. 2021; Niemitalo & Ekkerman, 2022). The chapter concludes by assessing the validity and reliability of the methodology. The results, answering the second and third research questions, are presented in the fourth chapter. This presentation includes the optimization trajectories of the optimized routings and their performance in both cases. Furthermore, the results are validated by benchmarking them against randomized and baseline routing. In the fifth chapter, the results are discussed, analyzed, and compared, leading to the presentation of the study's key findings. Finally, the sixth chapter concludes the study by addressing its theoretical and practical implications, assessing the quality and limitations of the study, and suggesting ideas for future research on the topic. 7 2 CIRCULAR BIOECONOMY This chapter reviews the literature regarding the circular bioeconomy. Firstly, the concepts of bioeconomy, circular economy (CE), and circular bioeconomy (CBE) are defined and distinguished. Secondly, the conceptual frameworks for CE and CBE are presented, discussed, and compared. In the third, fourth, and fifth sections of the chapter, the perspective of the literature review is taken to a more practical level by discussing and analyzing the implementation of circular business models (CBM) aligned with the principles of the circular bioeconomy in the context of the study’s case company, its supply chain, and the related success and risk factors. The chapter closes by synthesizing the literature review. 2.1 From bioeconomy and circular economy to circular bioeconomy The term bioeconomy refers to parts of the society and economy that utilize renewable biological resources from land and sea to produce materials, food, and energy. These parts of the society and the economy cover the sectors and associated services that process, produce or are in contact with any part of the supply chain involving biological resources. (Giampietro, 2019) One example of the bioeconomy is the utilization of agricultural waste and side-flows, such as manure and straw, as a resource in biogas production, in cooperation with actors from the energy and agriculture industries and their stakeholders. The term is intended to remind people of the biological origin of the economic process and highlight the problem of limited usable and accessible resources that are unevenly located and unequally appropriated (Giampietro, 2019). However, due to the renewable nature of bio-resources, they can be said to be naturally circular at some level (Salvador et al. 2021). The natural circularity of bio-resources may already enable greater environmental sustainability compared to practices using fossil resources (Paredes-Sánchez et al. 2019). Since the diversity of bioeconomies is greatly dependent on local characteristics and the possible scarcity of required resources, cooperation among the actors in the bioeconomy should be developed to enable more efficient utilization of circularity via the renewable nature of bio-resources. With more circular usage of bio-resources, economic growth and environmental benefits may be achieved by maximizing the value of resources and minimizing the consumption of virgin or non-renewable resources, approaching the principles of the CE (Salvador et al. 2021). 8 The CE has gained an increasing amount of attention and popularity during the last few years. This has led to a multitude of partly congruent and vague definitions of the CE by the science community (Korhonen et al. 2018; Prieto-Sandoval et al. 2018). Differing definitions of the CE have generated a need to define a general framework for it. It is important to understand the differences between the CE and sustainable development, whereas the principles of CE offer tools to attain sustainable development. Understanding the prevailing dynamics affecting circularity which are dependent on the level of perspective also plays a crucial role if sustainability is sought via circularity. Economic, environmental, and social dimensions of sustainability should also be considered to adjust the CE practices in a way that serves the purpose of reaching the selected type of sustainability. (Korhonen et al. 2018; Prieto-Sandoval et al. 2018) Stahel (2016), Korhonen et al. (2018) and Kirchherr et al. (2018) discuss the defining factors of the CE, its opportunities, and its restrictions. The main aim of the CE is to close loops in industrial ecosystems. This is achieved by repurposing goods near the end of their useful life, thus reducing waste. This approach profoundly shifts the economic paradigm from production to sufficiency. The focus is on deriving value from existing goods and manufacturing only when necessary. This fundamental departure from non-circular economies classifies CE business models into two non-exclusive groups. The first prioritizes reuse and prolonging product life through repairs and upgrades. The second group seeks value by transforming goods and materials into valuable resources through recycling. (Stahel, 2016) However, implementing circular practices and business models must address challenges arising from current industry practices to achieve sustainable development. (Korhonen et al. 2018; Kirchherr et al. 2018). The challenges of implementing CE practices arise from the physical scale of the economy and its impact on sustainability, as well as the path-dependency and lock-in effects (Korhonen et al. 2018; Kirchherr et al. 2018). Even if CE practices increase the economy's scale, harmful impacts can result (Korhonen et al. 2018). Jevons' paradox, described as lower costs leading to increased production and consumption, complicates matters (Alcott, 2005; Korhonen et al. 2018). Economic growth presents the "boomerang effect”, where pollution-intensive industries may relocate to poorer countries during sustainability efforts (Korhonen et al. 2018). Path-dependencies and lock-in challenges stem from market dynamics. Early solutions dominate, hindering new CE innovations. (Korhonen et al. 2018) Pricing dynamics of recycled materials are perceived to be more volatile in comparison with prices of virgin materials, influencing the consumption behavior of the markets (Grafström & Aasma, 9 2021). Kirchherr et al.'s (2018) study highlights barriers to CE: cultural (consumer disinterest, linear operations), market (low virgin material prices, high upfront costs), regulatory (policy framework gaps), and technological (lack of circular design and remanufacturing) barriers. Technological deficiency can be further extended to encompass a lack of technological capabilities. Businesses without suitable IT systems might find it difficult to evaluate costs and profits associated with circularity investments using their operational data (Grafström & Aasma, 2021). This can lead to unsupported decisions to either reject or accept investments in CE. While businesses prioritize cost and competition, policymakers recognize the importance of regulatory shifts toward supportive CE frameworks (Kirchherr et al. 2018). Circularity brings some additional benefits if its requirements are satisfied. Applying circular practices, such as recycling and reprocessing of materials, generates jobs and saves energy with lower resource consumption and waste (Stahel, 2016). By considering these kinds of indirect benefits of the CE, challenges for implementing CE practices profitably that are hard to affect directly, such as thermodynamic and system boundary limits (Korhonen et al. 2018), may be addressed indirectly. For instance, a car owner choosing between buying new tires or repairing old ones illustrates the circular ecosystem. Waste management services are crucial for this system to work, allowing used items to be resold rather than discarded. The main goal is profitable waste management to enhance circularity, achieved by minimizing collection and disposal costs. To optimize resource value over time, proper markets and collection points are needed for recycling, based on the material being considered. (Stahel, 2016) While defining the marketplaces and collection points that enable circularity, the cost caused by using the circular channel of recycling materials to individuals must be considered. For instance, one may not sort the waste he produces, if dumping all goods into mixed waste containers is perceived significantly more cost-efficient to the individual than sorting the waste. However, if sorting of waste is sufficiently encouraged through selected incentives and restrictions, the collection of certain types of waste, especially biobased waste, becomes possible. This enables the implementation of a circular bioeconomy and its opportunities. The circular bioeconomy (CBE) refers to the overlapping parts of the bioeconomy and the CE. One definition of the CBE states that its implementation enables sustainable economic growth in developed countries by considering the CE as what should be done, and the bioeconomy as how it should be done (Giampietro, 2019). The implementation of CBE promotes more sustainable manufacturing practices in general, leading to the transformation towards a broader usage of renewable bioresources, which are converted 10 into bioenergy and other bio-based products. This, in turn, generates new jobs and industries (European Commission, n.d., as cited in Giampietro, 2019). The CBE is a system that maximizes the utilization of bioresources in manufacturing and energy production, adds high sustainable value with the reuse and cascaded use of materials, and minimizes the virgin resource input from the natural environment, as well as the emissions and resource outputs to the environment. In comparison to the bioeconomy, the CBE aims to keep bioresources within the technical cycle for as long as possible. (Salvador et al. 2021) Therefore, the CBE aims to maximize the total value generated per unit of a resource rather than maximizing the value generated per resource unit in a fixed measure of time or usage. The approach of solely seeking the best and most effective economic impact in a fixed service life of a resource is an old-fashioned and less innovative approach, typically implemented in the context of the bioeconomy, not in the context of CBE (Salvador et al. 2021). 2.2 Conceptual framework for circular bioeconomy Prieto-Sandoval et al. (2018) propose a framework for the CE, highlighting its relationship with eco-innovation. In this section, the conceptual framework of CE is reviewed and the CBE’s positioning in it is examined. After that, the conceptual framework of CBE presented by Wei et al. (2022) is analyzed in detail and discussed and compared to the CE framework by discussing the similarities between the frameworks. Analysis of Prieto-Sandoval et al. (2018) suggests that the CE should include four main components: minimizing resource demand, a multi-level approach, importance for sustainable development, and the connection between the CE and innovation (Prieto-Sandoval et al. 2018). The first and third components align with Stahel's (2016) definition, emphasizing the minimization of resource consumption and the enablement of sustainable development. The multi-level approach and connection to eco-innovation offer new insights into understanding the CE. Yuan et al.'s (2006) research informed Prieto-Sandoval et al.'s (2018) introduction of three levels of the CE implementation: micro-, meso-, and macro-level. At the microlevel, companies focus on implementing CE practices that serve their own interests, such as reducing costs and enhancing their reputation (Prieto-Sandoval et al. 2018). While this may generate economic benefits, it is important for regulators to incentivize companies to operate with circular principles to ensure that CE practices are implemented for the right reasons. 11 At the meso-level of the CE, companies participate in industrial symbiosis, which benefits both the regional economy and the environment (Prieto-Sandoval et al. 2018). The focus is on encouraging the creating of eco-industrial parks and networks generating both types of benefits (Geng et al. 2012; Prieto-Sandoval et al. 2018). An example of a mesolevel network is a group of companies operating in different industries, whose operations are related in some way. Cooperation between these companies creates economic and environmental benefits, such as a biogas network producing and selling renewable energy. This reduces the need for non-renewable energy sources and provides a sustainable alternative. The macro-level of the CE involves focusing on cities, municipalities, provinces, and states, and their circularity. This includes developing eco-cities, eco-municipalities, and eco-provinces, which require institutions to raise awareness and set boundaries and incentives to implement the CE practices. (Yuan et al. 2006; Prieto-Sandoval et al. 2018) Raising awareness for the need to improve circularity affects the funding of research, leading to more sustainable development. This research provides opportunities for ecoinnovation, which is crucial for improving the CE practices and achieving sustainable development. Eco-innovations drive the improvement of existing practices across all three levels of the CE (Prieto-Sandoval et al. 2018). They are defined as improved products, services, processes, marketing changes, or organizational changes that decrease waste and emissions across the entire supply chain of the given context and reduce the use of natural/virgin resources (Hojnik & Ruzzier, 2016). Ultimately, eco-innovations address the environmental impact of innovation by promoting more efficient resource usage while simultaneously minimizing the harmful environmental impacts of current practices (Hojnik & Ruzzier, 2016). Three determinants of eco-innovations in the CE are identified: policy and regulation, supply side, and demand side. Policy and regulation determine the legal framework for the CE that should support circular supply side actions such as cleaner production. (Prieto-Sandoval et al. 2018) Incentives are also provided to execute sustainable action to gain additional benefits. Demand side determinants consist mainly of consumers, who should be able to acquire sustainable behavior and accept eco-innovative products in the market instead of less sustainable products (Prieto-Sandoval et al. 2018). Policy and regulation determinants may influence demand-side determinants through market dynamics, such as sales restrictions or taxes on non-sustainable products. Hidden determinants such as market-specific factors can also affect the capability of actors to adopt 12 sustainable behavior, such as the government's view on the importance of green transition. Acceptance of the importance of green transition can lead to funding research on the CE, enabling more eco-innovations to emerge. In more detail, eco-innovations may be divided into product, process, and organizational eco-innovations, which emerge from environmental R&D investments driven by factors such as regulatory requirements, investments, market pull, operational efficiency needs, and cost savings (Hojnik & Ruzzier, 2016). As a result, eco-innovations may not be solely technological; they can also encompass organizational, social, or institutional innovations. This is why they can emerge from either companies or nonprofit organizations, and they may not always be tradable in the markets. Due to the complex and multifaceted nature of eco-innovations, an interdisciplinary approach is usually required for their emergence. An interdisciplinary approach is necessary to understand the economic market factors that determine the success of innovation. On the other hand, the innovation itself must be sufficiently advanced to generate enough environmental value, such as waste reduction and minimized use of virgin materials. (Hojnik & Ruzzier, 2016) Since it is argued that the CE offers tools to achieve sustainability (Korhonen et al. 2018; Prieto-Sandoval et al. 2018), a similar argument can be made for the necessity of ecoinnovations to emerge to design and enhance the current CE practices. These practices enable more sustainable operations and, ultimately, sustainable development by consistently improving the adopted methods through eco-innovation. Sustainability is defined as the attainment of a well-coordinated balance among economic, social, and environmental performance across different generations (Geissdoerfer et al. 2017). This equilibrium involves respecting nature's capacity for self-regeneration by avoiding overconsumption and staying within the limits of nature's production capacity. The three pillars of sustainability – people, profit, and planet – encapsulate the idea of a necessary balance among environmental, economic, and social dimensions while striving for sustainability. These pillars aim to preserve the functions of the Earth's ecosystems, nurturing well-being, health, and security, all of which are crucial for ensuring a sustainable future. (Geissdoerfer et al. 2017) However, since sustainability may be perceived as a high-level concept, its set of goals, which involves pursuing simultaneous benefits for the environment, economy, and society (Geissdoerfer et al. 2017), is broader and less defined than the corresponding set of goals for the CE. On the other hand, the goals of CE are centered around closing the loops within material flow systems (Geissdoerfer et al. 2017). If the CE practices are considered as prerequisites for sustainability and similarly regard eco-innovations as prerequisites for the CE practices, it becomes essential to consider the factors that drive 13 and influence the adoption of eco-innovations (Bossle et al. 2016; Hojnik & Ruzzier 2016). These forces driving the success of eco-innovations may ultimately determine whether sustainable development is achieved or not. The widespread adoption of eco-innovation requires innovation itself to emerge and circumstances to adjust for broad adoption across actors in the examined market or industry. This entails considering established factors that impact eco-innovation adoption at the company level (Bossle et al. 2016; Hojnik & Ruzzier 2016). Eco-innovation's emergence is primarily driven by regulations and market pull factors, which are crucial for its emergence within companies (Hojnik & Ruzzier, 2016). These factors, aside from the type of eco-innovation, play a vital role in its development, adoption, and diffusion phases (Hojnik & Ruzzier, 2016). Factors influencing the adoption of eco-innovations and achieving sustainable development can be categorized as internal and external factors within an organization (Bossle et al. 2016). Regulatory and market demands also impact eco-innovation adoption, compelling organizations to embrace new practices, resources, or processes. For instance, customers and stakeholders may demand operations that align with corporate social responsibility (CSR), which can be enhanced by impactful eco-innovations. Internal factors affecting eco-innovation adoption at the company level encompass environmental capabilities, strategy, human resources, and the drive for operational efficiency. Industry sector and company size similarly influence the adoption process. (Bossle et al. 2016) Based on the previously presented review and analysis on the CE, an illustration of its central concepts is presented on the Figure 2. The illustration includes the main process of the CE representing the ideas of circular material flows and the crucial role of innovation in enabling the implementation and development of the CE practices. Subsequent and key social, environmental, and economic benefits are also illustrated and emphasized. See page 14 for the Figure 2. The positioning of the CBE in the context of the CE will be discussed next. 14 Figure 2. Conceptual model for the circular economy (CE). Graph adapted from Stahel (2016) and Korhonen et al. (2018). 15 Since the CBE is a sub-concept of the CE that focuses on biological materials (Giampietro, 2019), it can be argued that the framework presented in Figure 2 for the CE is still applicable when discussing the CBE. However, one must consider the potential boundary conditions established by the CBE. For example, the variety of extracted resources significantly decreases when focusing on the CBE rather than the CE. A common mechanism used in the CE manufacturing industries to repair goods simply disappears in the CBE. While inspected materials have processing value in the terms of energy and nutrients, there are no logical ways to repair or refurbish them, limiting the ways in which the materials can be used compared to the CE. Additional examples could be easily generated to highlight the differences between the CBE and the CE. Nevertheless, in the context of Figure 2, especially its inner circle that represents the material flow in a circular system, the group of alternative ways to execute operations in each phase (use, innovation, extracted resources, manufacturing, distribution, and use) is significantly narrowed when the focus shifts from the CE to the CBE. Thus, one should consider the CBEspecial boundary conditions and restrictions caused by the focus of biological materials. The CBE is argued to bring environmental and economic benefits by Mohan et al. (2018) and Salvador et al. (2021). The benefits are achieved via the recovery of bio-based wastes and byproducts, preventing pollution, and promoting valorization, allowing the sale of wastes as marketable products with added value, thus enabling economic growth (Salvador et al. 2021). It can also be argued that the CBE brings social benefits, but not as much as the CE in general. The CBE's contribution to generating jobs and industries in capturing the value of bioresources is undisputed, whereas the complete supply chain of biomass procession generates a variety of jobs in capturing the value. However, the other typical aspects of the CE's social benefits may not be as predominant in the context of the CBE. The sense of community, cooperation, and sharing functions and services are probably less relevant in the CBE since the boundary conditions of the CBE do not naturally encourage the sharing economy as strongly as the CE in general. Reuse, refurbishment, and other measures that encourage the sharing of recycled goods are generally more applicable in the CE. 22 Based on the inefficiencies in the AD process (where not 100% of the biomasses are degraded), it could be reasonable to ask whether it would be more profitable and desirable to sell and use the manures generated on the farms as fertilizer in the first place, instead of first transporting it in and out of the biogas facility? This question becomes relevant when the supply of manure and the demand for manure as a fertilizer are in balance. However, in most cases, the availability of manure is usually higher than the fertilizer needs (Kamilaris & Prenefeta-Boldú, 2021). In addition, it has been found that utilizing surpluses of digestates as fertilizers leads to a lower level of emissions compared to spreading raw manure on fields as a fertilizer (Jensen et al. 2017). Therefore, utilizing digestates as fertilizers instead of raw manures enhances the implementation of the CBE principles through decreased emissions. The type of manure being inspected here is also a relevant question since research results have concluded that the profitability of setting up treatment units to compost manure for later fertilizer use depends on the species of the manure being inspected. Another factor to consider is the season of the review period, as there may be national policies restricting the use of solid manure as a fertilizer during certain times. (Kamilaris & Prenefeta-Boldú, 2021) However, it should also be considered that transporting manures to be used as fertilizer comes with considerable logistic and transportation costs as well, although their usage decreases the need to use artificial fertilizer inputs (Lessman et al. 2023). It has also been found that the number of harmful pathogens in digestate resulting from the AD process is linked to storage time and other process parameters, such as temperature and the carbon-nitrogen ratio of the feedstock-manure (Nag et al. 2019). There is no reason to question whether pathogens also exist in pure manure, which reduces the attractiveness of both digestate and pure manure for use as fertilizer in the food industry, since the usage of organic materials as fertilizers can contribute to pollution (Lessmann et al. 2023). From this perspective, the most reasonable solution to this question would be to produce as much biogas as possible and, depending on the AD rate of the process, decide whether transporting the resulting digestate is reasonable and cost-effective. The emergence of harmful pathogens should be ensured to be minimized by considering the factors that affect them (Nag et al. 2019). Therefore, by reviewing the economic and environmental costs and benefits of transporting the digestates back to farms, the case company could increase the probability of aligning its operations with the practices of CBE. 23 2.4 Supply chain of the biogas production In Figure 4, an example of a circular supply chain is depicted, which bears resemblance to the supply chain employed by the case company. The illustration showcases the utilization of CBMs within this circular supply chain. See page 24 for the illustration. Following that, this section examines the phases occurring in the supply chain and the related terms presented in Figure 4 in detail. 24 Figure 4. Example illustration of the case company’s circular supply chain. For circular bioeconomy supply chain activities, see Ponjavic et al. (2020). For more details about the selected customer, see Jumppanen (2023). 25 Figure 4 presents the value creation process and supply chain of biomasses initially generated on farms. They are then transported to be used as feedstocks in biogas production, followed by the value realization through profits made from selling the biogas and digestates as fertilizers. The process phases are described as self-explanatory in Figure 4. However, the phases occurring at the biogas facility, which cover the warehouse processes, material flows, and actual biogas production, include certain indicator terms that measure the performance of those phases within the supply chain. The meaning of these indicator terms should be explained. The indicator terms are inventory turnover, delivery time, order point, EOQ, production capacity, MRP, production planning, EPQ, and quality management. These terms presented above are all closely related to the philosophy of lean production (Demeter & Matyusz, 2011). Lean aims to satisfy customer demand at the highest possible level while minimizing wastes, losses, and other inefficiencies throughout the supply chain. If executed effectively, lean procedures enable lower storage levels and faster inventory turnover, thereby reducing storage costs and waste, and therefore aligning practices with the principles of CBE. (Demeter & Matyusz, 2011) Since the demand for biogas is assumed to be relatively steady in the case company's operating environment due to the chosen strategic customer (Jumppanen, 2023), and the product portfolio of the biogas plant is narrow, consisting of biogas and digestates sold as fertilizers, the benefits of lean are more achievable, according to Demeter & Matyusz (2011). These benefits within the operation of the supply chain can be achieved by the case company by making wise choices of biomass feedstock suppliers, preferring suppliers with high delivery security and biomass production capacity, as well as aiming for short and predictable delivery times. However, since real-world supply chain processes are not deterministic and various disruptions may occur within them, planning and operating the supply chain for the biogas production should consider being prepared for feedstock supply disruptions, biogas production stoppages caused by feedstock shortages, and other factors that may lead to production interruptions. Additionally, uncertainty in biogas and fertilizer demand should also be taken into consideration. To prepare for feedstock supply disruptions, it could be beneficial to employ biomass type-specific reorder points and order quantities, ensuring a continuous production process through the use of safety stocks (Sevgen & Sargut, 2019). The determination of order quantities and reorder points should be guided by a model that takes into account case-specific uncertainties in relevant factors, such as demand levels and lead times (Chaharsooghi & Heydari, 2010). 26 The reorder point establishes a threshold for the storage level of a feedstock, triggering the creation of a new purchase order when the storage level falls below it. When the storage level is below the reorder point, a purchase order with the specified quantity for the feedstock is sent to the supplier. By considering the costs of purchasing, including fixed and logistic costs, storage costs, and the expected demand for the feedstock, the economic order quantity (EOQ) can be defined. This helps determine the optimal order quantity for purchases. (Sevgen & Sargut, 2019) If an optimal order quantity can be defined and utilized, waste, losses, and costs caused by procurement operations could be reduced, thereby enhancing compliance with the principles of CBE. However, the efficient utilization of optimized values for order points and quantities requires commitment from all parties across the supply and demand sides (Chaharsooghi & Heydari, 2010). To ensure the commitment of supply-side actors, models for reorder points and quantities that also incentivize suppliers could be developed. These models incentivize actors on the supply side by providing shared economic benefits through commonly agreedupon agreements and practices related to the procurement and supply of feedstocks. (Chaharsooghi & Heydari, 2010) Given that the demand for biomass feedstocks is cyclical and discontinuous due to the long processing time of a feedstock batch within anaerobic digestion (Su et al. 2022), defining and applying reorder points and order quantities for biomass feedstocks could significantly help the case company prepare for feedstock shortages without interrupting biogas production. However, higher storage levels may lead to additional storage costs and feedstock spoilage, reducing the biogas and fertilizer yield of the production process (Hadin et al. 2016). Therefore, the reorder points and order quantities should be defined with careful consideration of all other possible and relevant consequences. This includes determining threshold values for potential acceptable spoilage rate and storage costs. Although the demand for the biogas produced in the case company's biogas reactor can be assumed to be rather steady and predictable due to the chosen strategic customer (Jumppanen, 2023), unpredictable disruptions may still occur, impacting biogas demand. It should be noted that these potential disruptive impacts on demand, whether negative or positive, should be considered. Regardless, preparation for demand disruptions can be achieved by establishing and maintaining biogas storage facilities near the biogas plant (Jensen et al. 2017; Butemann & Schimmelpfeng, 2020). Since biogas is unlikely to spoil like feedstock manures, it is reasonable to suggest that the case company prioritize biogas production over storing manures and straws when biogas production is not in progress. This suggestion holds if the costs of biogas and 27 fertilizer storage are reasonable and acceptable (Jensen et al. 2017). However, it is important to consider that the unit storage cost for biogas may be higher compared to the unit storage cost of feedstocks, as gas has a lower density than feedstocks, requiring more storage space and containers. Once again, the actual decision regarding the prioritization of storing gas and fertilizers with continuous biogas production instead of intentionally maintaining feedstock storage levels should always be made by considering all potential consequences related to total costs, emissions, and biogas and fertilizer nutrient yields. These decisions should align with the case company's production planning and material requirements planning (MRP) processes, which determine the biogas and fertilizer production batch sizes and schedules to meet the expected demand (Stevenson, 2014, pp. 495–515). MRP is a process carried out in conjunction with production planning, where the material requirements for planned production volume are determined, and upcoming purchase orders are scheduled. Production capacity should be considered alongside production planning and MRP, since it may restrict the production volumes in the time unit, which may lead to oversized feedstock storages, causing pollution and waste. (Stevenson, 2014, pp. 495–515) Differences in biogas yield potential across biomasses to be utilized should also be considered when defining the composition of biomasses for AD. For instance, manures have a lower biogas yield compared to straws (Jensen et al. 2017), indicating that the proportion of manure in the composition should be restricted to achieve the planned biogas production volumes. Alongside material requirements and production planning, the economic production quantity (EPQ) could be determined and utilized to define an economically optimal production batch size for biogas, as well as for digestate as a by-product (Karmakar et al. 2017). EPQ is fundamentally similar to EOQ (Sevgen & Sargut, 2019), which takes into account inventory and fixed costs of production, as well as the demand for biogas and fertilizers. It proposes an optimal batch size for biogas and digestate production. If executed with justification and efficiency, an optimal production schedule resulting from production planning and MRP processes could ultimately improve the economic viability of the biogas plant through decreased production costs (Jensen et al. 2017). This would enable more resource allocation towards improving the practices that enhance CBE. Alongside production planning, potential incentive structures for the biogas production set by the regulators should also be considered if they exist. If the incentive structure enables it, it may be reasonable and economically more viable to schedule production of biogas for times of high market prices for electricity, while favoring the storage of biogas and fertilizers during times of low market prices (Butemann & Schimmelpfeng, 2020). 28 This would be the case for the case company's biogas plant if they decided to let markets price biogas and fertilizers to be delivered. However, due to the chosen strategic customer (Jumppanen, 2023), the prices for biogas are probably set by a mutual agreement. Additional costs caused by recurrent stopping and starting combined heat and power units would result in avoidable costs and wear and tear (Butemann & Schimmelpfeng, 2020), which may decrease profits and unnecessarily commit critical resources of the biogas plant, instead of using those resources to improve the practices enhancing CBE. However, if incentive structures of this kind are found to exist, and the biogas and fertilizers are sold at market prices, the case company should consider this as a relevant factor while planning its biogas production, since it is possible to increase revenues and minimize additional costs caused by the flexible production of biogas (Butemann & Schimmelpfeng, 2020). The last concept to be explained is quality management. Although it is illustrated as being related to the warehouse and production phases of the supply chain in Figure 4, it should be noted that quality management practices should be considered and executed throughout all phases of the supply chain. In the context of processing biomass materials consisting of manures, grass, and straws, the quality management practices to be implemented should focus on the inactivation of harmful pathogens within the manures and digestates (Nag et al. 2019). If harmful pathogens still exist in digestate to be reused as fertilizer, the spreading of serious illnesses may follow as a consequence, especially if digestates are used as a fertilizer in industries producing food for humans and animals (Liu et al. 2008; Kinyua et al. 2016). The pathogens have the potential to spread through air, water, and direct contact, which highlights the need for careful treatment of manures and digestates. (Kinyua et al. 2016; Nag et al. 2019) However, there are ways to increase pathogen inactivation throughout the AD process by setting up process parameters to create conditions that increase the probability of inactivation (Nag et al. 2019). High process temperature significantly contributes to pathogen inactivation (Manyi-Loh et al. 2013; Nag et al. 2019). The biogas production lead time, or hydraulic retention time (HRT) in other words, also has an impact on pathogen inactivation. Depending on the composition of the biomass to be digested, an HRT of 12 to 35 days could contribute the most to pathogen inactivation as well as biogas yield. Other important factors to consider in decreasing the pathogen count are maintaining a high organic loading rate (OLR) throughout the AD process and pasteurization of materials to be processed. (Sakar et al. 2009; Nag et al. 2019) 29 2.5 Success and risk factors of the biogas supply chain The implementation of CBMs within the case company's supply chain involves additional success and risk factors worth noting. Donner et al. (2021) analyzed critical success and risk factors for CBMs that valorize agricultural biomasses and side-flows. From the perspectives of logistic optimization and AD, it should be noted that the outputs of the biogas plant and the efficiency of the logistic operations may be influenced by the need to transport and process different types of biomasses and by-products (Donner et al. 2021). From a logistic perspective, this is critically important as it can impose restrictions on transporting certain materials together, leading to significant impacts on logistics planning and costs. Depending on the TS rate within the plant’s storage and production, production may be disrupted if water dilutions are required to reach TS lower than 15 %, decreasing the efficiency of biogas production and restricting it from reaching its full potential. Thus, co-transporting of materials should always be utilized when required and possible from the perspectives of logistics and production management, but also in terms of emissions. It was found by Yu et al. (2023) that co-distribution of straw and manure contributes most to GHG mitigation. The processing capacity of the biogas plant should also be balanced with the material input from the logistic system. As noted by Donner et al. (2021), high storage capacity for feedstock located nearby the AD processing system may serve as a critical success factor if the biomass to be processed can be stored. However, this incurs storage costs for the case company, requiring analysis and optimization of the balance between storage capacity and delivery times within the logistic system. It is important to consider the possible time sensitivity of the quality of biomasses to be stored, as their accumulation can lead to biomass spoilage and decrease the facility's AD-rate. From this, it can be concluded that with an optimal solution to the logistic routing problem, the utilization rates of biomasses can also be improved, as their quality in terms of methane potential and nutrient content is assumed to be a time-critical variable, as found in the case of horse manure by Hadin et al. (2016). This is intuitive, since if the quality decreases over time, so does the amount of material to be valorized. This means that with an optimal solution to the logistic problem, pollution of materials can also be decreased. When balancing production capacity, logistics and warehouse management, it should be also noted that the availability of side-streams valorized through AD can be affected by seasonality (Donner et al. 2021). Thus, depending on the magnitude of technological investments in production capacity and valorization rates of processed materials, it should be reviewed what types of materials it is most reasonable to valorize in different seasons. A probable suggestion would be to valorize only manures during the winter and 30 inspect the costs and benefits of valorizing straws alongside manures during the summer. On the other hand, since the grass and straw are baled in the fields, their collection is permitted throughout the year, although they are accumulated only during the summer. However, making valid recommendations to utilize only a selected group of biomasses in specific periods would require further and more detailed analysis, which is excluded from this study. The collection and utilization of all biomasses of interest are assumed to be continuous, although the peculiarities in their accumulation dynamics are taken into consideration. The geographical distribution of biomasses and side-streams to be picked up should also be noted as planning the logistic operations. Agricultural wastes and side-streams have varying quality and limited volume depending on the farm (Donner et al. 2021), making it more appealing to pick up biomasses from specific farms. Therefore, farms with high quality and volume of biomasses should be recognized and establish strategic partnerships with their owners. If the required biomass is picked up from farms with high volume and quality, it follows that the need for logistic operators to pick up biomass from several farms decreases. Thus, it can be reasoned that the total mileage driven by the logistic operators decreases, indicating lower fuel consumption and emissions caused by the vehicles. Efficient resource consumption and the minimization of emissions should remain as key design principles when planning logistic operations. This is because the total contribution of bioenergy production systems to the ecological footprint has been questioned by Wang et al. (2020), who concluded that biomass-based energy production increased the ecological footprint of G7 countries. Therefore, aligning practices with CBE principles, such as minimizing emissions and maximizing resource usage, should not be taken for granted. Instead, the pursuit of these practices should guide all related decision-making processes in planning the logistic operations as well. 2.6 Synthesis of literature review Based on the previously presented literature review, it can be generally stated that the profitable and efficient utilization of CBMs requires in-depth analysis of the prevailing market, industry, and company-related factors and dynamics. Due to the broad set of ways to minimize resource consumption and maximize their utilization, the material flows and the processes affected by them should be defined and well-known before choosing the parameters of the CBM to be utilized. However, in the context of the case company and the CBE, it seems that the selection of actors and processes in the supply chain and processing technologies to valorize biomass materials through bioenergy production are justified. These selections aim to 31 achieve aligning practices with the principles of CBE, which include reducing emissions, minimizing costs, maximizing biomass utilization, and enhancing sustainability by proposing bio-based energy alternatives. In planning the logistic operations related to transporting, storing, and utilizing biomasses such as manures, grasses and straws collected from nearby farms, the case company should optimize its supply chain. This optimization involves optimizing the transportation of biomasses from farms to the biogas plant, as well as warehouse and production management. The optimization of transportation logistics for biomasses is thoroughly examined in this study. Additionally, the examination covers monitoring stock levels of feedstocks and their consumption within the biogas production process. However, other factors related to warehouse management and production control are excluded from the analysis. The costs caused by the logistic system may have subsequent implications for endangering the achievement of aligning practices with the principles of CBE by consuming the company's resources, namely money, on unproductive operations such as logistics. These resources could be better invested in funding research on more advanced bioprocessing techniques and practices, ultimately enhancing circularity itself. Based on the literature review and synthesis, a theoretical framework was designed to highlight the factors of the biogas supply chain that need to be improved and can be enhanced with developed data processing and dynamic optimization tools. These tools will be examined in the following chapter, and the framework is presented below. Figure 5. Framework of the factors within the biogas supply chain that can be improved with developed methods. 38 costs are incurred. Similarly, if the total amount of biomass imported during the year surpasses the targeted annual input, additional costs are incurred in the form of unnecessary imports. The dynamic accumulation of biomasses at the sites transforms the BRBPP into a dynamic optimization problem (Liu et al. 2019). For instance, a decision to visit a certain site a week later instead of immediately affects the total costs of a solution, as this decision could ultimately impact the occurrence of a production stoppage at the biogas plant if storage runs out. This is due to the choice of early site visit with a lower level of biomass to be collected, rather than visiting a site with a greater level of biomass and postponing the visit to the initial site for the following week. Additional storage costs caused by exceeding the site’s storage capacity could also arise, depending on whether the initial site was visited immediately or not. The illustration of the BRBPP is presented in Figure 6; refer to page 39 for that. The illustration highlights various types of pickup sites and vehicles. The scenario depicted in the illustration can be considered a starting point for a typical day within the simulation period, providing an example of a partial solution to BRBPP. At the beginning of each day, every vehicle is assigned a route, which is essentially a list of locations to be visited throughout that day. A vehicle's route may also be empty, indicating that no transportation is required for that vehicle during that day. The solution encompasses routing information for each vehicle for every day of the simulation period. For example, if the simulation period is set to last 251 days and routing is calculated for 9 vehicles, the BRBPP solution includes routing details for those 9 vehicles over the coming 251 days. The solution is derived through optimization facilitated by the GA, which simulates the system with generated routing proposals and calculates the costs incurred with the given routing. The cost function is shown in the lower right corner of Figure 6; a more comprehensive analysis of it is provided in Chapter 3.5.2. 39 Figure 6. An example illustration of a partial solution to the Biomass Routing and Biogas Production Problem (BRBPP). 40 As can be detected from the cost function presented in the Figure 6, the value of a cost function is determined by multiple variables. The presence of multiple variables affecting to the costs of a solution proposal alters the BRBPP multi-dimensional optimization problem, whose solutions can be sought by multi-objective optimization methods (Jozefowiez et al. 2008; Baños et al. 2013; Banasik et al. 2017). The difference is fundamental in comparison with the classical VRP, whose solutions minimize the total distance driven by the operators with the requirement of visiting each node once, thus being a singleobjective optimization problem (Jozefowiez et al. 2008; Baños et al. 2013). By incorporating multiple cost terms, which are objective variables to be minimized through optimized routing, the optimization of biomass side-flows within a circular supply chain allows us to discover more comprehensive solutions. These solutions consider the relevant factors that affect costs in the system, thereby enhancing the overall efficiency of the supply chain. (Jozefowiez et al. 2008; Baños et al. 2013; Banasik et al. 2017). The cost function presented in Figure 6, examined in detail in Chapter 3.5.2, aims to find a logistical routing for a simulation period that optimizes the operation of the biogas production supply chain, from the farms to the biogas reactor. This optimization should mitigate production interruptions in the biogas production due to resource shortages, minimize the consumption of dilution water in production, prevent overflows and excessive visits to the facility. This can be achieved through route planning that simultaneously minimizes driving distances (and wrong visits), reduces overtime for drivers, and ensures sufficient storage capacity on a farm-specific basis for biomass production. 3.3 Data collection and description The data used in this study was collected from LuKe's Biomassa-atlas (LuKe, 2023). The Biomassa-atlas provides geospatial datasets for various types of biomasses located in Finland. The methodology used in creating and calculating Biomassa-atlas’s datasets is described in detail by Luostarinen et al. (2017). The datasets include information on the year of dataset collection, biomass type, mass in tons, and location coordinates for each biomass point. Location coordinates for each mass point were given as multipolygons, including corner points of the area covering the biomass. The coordinates were initially given in the EUREF-FIN coordinate system and were later transformed to the WGS84 coordinate system as part of the data processing. These corner points were used to calculate an average point of areas. Hence, average points were approximated to include an area’s mass, to enable processing biomasses as unique points. An example of the unprocessed data is illustrated in Figure 7. 41 Figure 7. An example of the form of unprocessed data. All datasets collected and used in this study were initially in a consistent, unified format, as illustrated previously. However, the datasets related to manures were on the municipal level, while the datasets for grass and straw were more detailed. The selection of datasets was conducted with case-by-case consideration to align the types of biomasses inspected in the datasets with the case company's intentions regarding the types of biomasses to be collected and used for the biogas production. The opinions of LuKe's experts were also taken into account regarding the types of biomasses that the case company is likely willing and able to utilize. The datasets chosen, processed, and utilized in this study included slurry manure of bovines and pigs stored in animal shelters in the years 2015 and 2016, dry manure of horses, bovines, poultry, and pigs stored in storage facilities in 2016, and side-streams of straw, grass, and silage grass generated in 2021. 3.4 Data procession The data processing tool was developed and written in Python as part of this study. The source code for the tool is publicly available (Eloranta, 2023a). The tool was developed to map the biomasses located nearby the biogas plant, determining the extent to which biomass potentials from surrounding areas could meet the resource needs of the plant’s annual production capacity. Additionally, the tool aimed to identify and approximate potential and optimal pick-up sites for logistic operators, from which biomasses could be collected and transported to the plant for use in biogas production. The pre-processed data was initially plotted to validate its correctness, including loading, transforming, and averaging operations. Figure 8 illustrates an example of this phase, where the xand y-axes represent the coordinates of the biomass points. The points are colored based on their mass, highlighting areas with higher biomass density. The color bar on the right side of the diagram represents the scaling system for coloring the points. 42 Figure 8. An example of the data pre-processing phase. After data validation, the datasets were filtered to include only biomasses located within a 50-kilometer radius from the plant’s location. The cutoff value for the 50-kilometer radius was set because it appeared that biomasses within this range could easily meet the plant’s resource demand. Expert opinions from LuKe and HAMK were also considered, along with the case company's interests regarding the resource demand levels. Euclidean distances were calculated for each point, and biomasses that were further than the chosen cutoff value were filtered out from the datasets. The accuracy of the distance calculations and filtering operations was validated by plotting the remaining data once again. An example of this phase of data processing is illustrated in Figure 9 below. 43 Figure 9. An example of the filtering phase in data processing. The remaining parts of the datasets were subjected to the unsupervised learning method, the KMeans clustering algorithm, to identify reasonable subareas (Jeffares, 2019; Magiya, 2019; scikit-learn developers, 2023). Approximations from these subareas were later modeled as pick-up sites in logistic simulation and optimization. The uneven geographical distribution of biomasses was taken into account during clustering by duplicating biomass points based on their masses. Mass coefficients were derived accordingly for duplicating points. This duplication was necessary because Sklearn's KMeans clustering algorithm does not consider weights in its clustering; it only considers coordinates without weight coefficients. The k value used for clustering was validated using the Elbow Method (Saji, 2023). The idea behind the Elbow Method is to cluster the dataset with varying numbers of cluster centers (k) and find the optimal value for k that minimizes the WCSS (within-cluster sum of squares). The optimal k is a value that is small enough to enable efficient calculations, yet large enough that further increasing k would not significantly minimize the WCSS (Saji, 2023). By plotting the WCSS as a function of k, a figure in the form of an elbow is formed, and the optimal k can be located at the elbow. An example of this validation is 44 presented in Figure 10, which was generated using the same dataset as the previous examples. Figure 10. An example of the Elbow method in validating the number of clusters. As can be seen in Figure 10, the WCSS has dramatically decreased as the number of clusters reached 25, approaching almost zero with 100 clusters. However, since these clusters will later be modeled as biomass pick-up sites in logistic simulation (i.e., farms), and the data being used is not farm-specific, it is important to choose the number of clusters cautiously. Additionally, it can be assumed that within a 50 km radius of the biogas facility located in Kanta-Häme, there are more than 25 farms. To ensure that the number of clusters more realistically represents the number of farms in the relevant area, the k value was initially set to 200 for clustering. This decision results in heavier computation but provides a more realistic modeling scheme with a very low WCSS value. An example of the initial clustering of biomass data within a 50 km radius, using 200 clusters, is presented in Figure 11. It is important to note that the unit of mass was converted to truckloads, assuming that 45 tons are equivalent to one truckload. Additionally, the masses were transformed to a logarithmic scale to enhance the visualization of biomass potentials for the case company. This allows for a better understanding of where the biomasses should be collected and from how far away to meet the resource demand. In Figure 11, each point located within a colorized sub-area represents the weighted central point of that sub-area, which includes all points related to the same cluster. Thus, the colorized sub-areas represent clusters in Figure 11’s visualization. The intensity of a color of the central point indicates the total amount of biomass within the cluster area. 45 These kinds of points will be modeled as pick-up sites, assuming that farmers within a cluster area will collect the biomasses and deliver them to the pick-up site for the logistic operator to retrieve. Figure 11. An example of the initial clustering phase in data processing. After the initial clustering presented in Figure 11, it was noted that the biomass resource demands for the case company's plant could be easily satisfied within a 50 km radius. The case company set the target input value for grasses and straws at 28,000 tons per year (Tampio, 2023), and it can be seen from Figure 11 that the biomass potentials of straws alone could theoretically meet the requirements. Similarly, with manures, the target value was set at 14,000 tons (Tampio, 2023), leading to similar conclusions. Due to the abundant occurrence of biomasses within the 50 km radius, the number of clusters was reduced to enable more efficient computation later on. It can be intuitively concluded that collecting biomasses as close to the facility as possible would be more cost-efficient, as it would reduce the total kilometers driven. Based on these observations and conclusions, a methodology for thinning the clusters was developed and implemented. 46 The thinning methodology utilized divides the initially clustered dataset of a 50 km radius area into 5 tires, with each tire having a thickness of 10 km. Then, the new desired value for k within each tire was calculated by dividing the number of clusters originally in the tire by a divisor coefficient raised to the power of the tire’s sequence number. With the new value of k for each tire, the biomasses within the tire’s area were clustered again, resulting in a lower total number of clusters. As a result, the number of clusters near the plant were thinned the least, while those furthest away were thinned the most. An example of the results of the thinning method described here is presented in Figure 12 below. Figure 12. An example of the methodology of cluster thinning. After thinning the clusters, the number of clusters was further reduced to enable more efficient computation. It appeared that the combined masses of different biomasses would exceed the targeted input values of the case company based on their resource demand. In this phase, the number of clusters was reduced by identifying the nearest clusters to meet the yearly biomass resource demand. Clusters that remained above the resource needs were filtered from the data. This filtering procedure was implemented for all biomass datasets, resulting in a significantly lower number of clusters to be calculated. At the same time, it ensured the adequacy of biomasses for the demand, considering 47 multiple grass, straw, and manure datasets. An example of the resulting clusters from this phase of data processing is presented in Figure 13 in page 48 in unified form with previous figures, and in Figure 14 in format of plotting cluster points within actual map format after transporting datapoints to GEOJSON format. After calculating the nearest clusters for each biomass type, the resulting datasets were combined to be used in simulation and optimization. It should be noted that the resulting datasets included information about the biomass type, thus ensuring that no information was lost when combining the datasets. The combination of datasets is visualized in Figure 15 in page 49, where the color of each point represents the biomass type. 54 was assumed that farms perform three cuttings during the summer: the first in mid-June, the second at the beginning of August, and the third in mid-September. Once these cuts are completed, the storage level at a pickup site goes from zero to one-third of the annual production for that site. To prevent simultaneous jumps at all sites, the jumps were distributed over three two-week periods: mid-June, the beginning of August, and mid-September. Each site was allowed a jump only once during a two-week period, resulting in a maximum of three jumps per site throughout the simulation period. During a two-week period, the accumulation process was modeled as follows: ∆𝑚𝑡,𝑖 =1 3𝑀𝑖𝜀𝑡,𝑖, (2) in which ∆𝑚𝑡,𝑖 is the accumulation of grass or straw at the day 𝑡 on the pickup site 𝑖, 𝑀𝑖 is the annual accumulation of grass or straw at site 𝑖, and 𝜀 is a vector consisting of 10 elements, with 9 of them being zeros and one element equal to 1. The location of this element within the vector, i.e., the date of the jump occurrence, was chosen randomly. The accumulation of grasses and straws follows Equation 2 only if the simulation is in one of the three two-week periods during which the cuttings are expected to occur. Otherwise, the accumulation of grasses and straws is zero. It should be noted that each site had a unique ε for each two-week period, ensuring that the cuttings did not occur on the exact same day for every site within that period, thus increasing the randomness of the simulation. Additionally, ε was initially generated for routing optimization and remained fixed for the optimization process. The decision to fix ε values was made because optimizing the routing would not yield actual optimality if ε values were completely random. This is because the optimization problem would change with each simulation run during the optimization process. To address the third research question, additional time-sensitive factors that influence the levels of biomass volume and moisture were incorporated into the model. The effect of passive drying of biomass was modeled, utilizing the storage model presented by van Dyken et al. (2010). The passive drying dynamics were applied to all simulation schemes. The rate of biomass volume loss due to passive drying is assumed to be 1% per week within the storage model of van Dyken et al. (2010). Therefore, the time-critical factor representing the volume losses of biomasses caused by passive drying was modeled as follows: 𝑉𝑏,𝑤 =𝑉𝑏(1 − 𝑟𝑙,𝑑)𝑤, (3) 55 where 𝑉𝑏,𝑤 is the volume of biomass 𝑏 at week 𝑤, 𝑉𝑏 is the initial volume level of biomass at the location of interest, and 𝑟𝑙,𝑑 is the rate of biomass volume loss caused by passive drying on a weekly basis. Due to passive drying, the moisture level of biomasses decreases, affecting their TSrate. The decreasing drying rate of biomasses was defined in timestep within the storage model of van Dyken et al. (2010). However, due to the broader complexity of this study's problem, including the accumulation and consumption of biomasses, the modeling of changes within the biomasses' level of moisture was simplified and not defined by timestep. In this study, a 5% decrease in moisture content per week through passive drying was assumed, based on the approximation of van Dyken et al.'s (2010) drying rate, which is dependent on storage time. Therefore, the time-critical factor representing the changes in the moisture level, later being transformed to changes within the TS rate, was modeled as follows: 𝑀𝑏,𝑤 =𝑀𝑏(1 − 𝑟𝑚,𝑑)𝑤, (4) where 𝑀𝑏,𝑤 is the level of moisture of biomass 𝑏 at week 𝑤, 𝑀𝑏is the initial level of moisture of biomass at the location of interest, and 𝑟𝑚,𝑑 is the rate of biomass moisture decrease caused by passive drying on a weekly basis. The dynamics of passive drying, as presented in Equations 3 and 4, were assumed to be valid at storages of pickup sites, during transportation, and at the storage of the biogas plant before being utilized within the AD process. The assumption of a constant pickup duration (Niemitalo & Ekkerman, 2022) was rejected within this study. This rejection may be justified by the coexistence of multiple types of biomasses and the dependence of pickup duration on the amount of biomass to be picked up. It was assumed that the pickup duration consists of a constant factor and a variable that is linearly dependent on the amount of biomass to be collected. The pickup duration may be formatted as follows: ∆𝑡𝑖,𝑙 = 𝑇+ 𝑣𝑚𝑙 , (5) where ∆𝑡𝑖,𝑙 is the pickup duration for the pickup site 𝑖 with a load level of 𝑙, 𝑇 is the constant term of the pickup operation, and 𝑣𝑚 is the collection rate for the biomass 𝑚. 𝑇 was assumed to be 10 minutes for all biomasses, based on the discussions had with the experts of HAMK and LuKe. Therefore, T may be considered as the setting time of a pickup operation, including the time spent on setting up and dismantling the collection equipment. Based on the discussions with the experts, 𝑣𝑚 was assumed to be 1.6 tons/min  0.625 min/ton for slurry manures, 1 ton/min  1 min/ton for dry manures, 56 and 1.2 ton/min  0.833 min/ton for grasses and straws. Although grasses and straws accumulate only in summer, it was assumed that once the cuttings are done, they are baled and stored in the fields, allowing the collection of them to be continuous throughout the year. It should be noted that the vehicle may not be able to empty the site if its load reaches its capacity. In that case, the vehicle loads only the amount it′s able to, and the time spent on that operation is calculated based on the loaded level. To facilitate monitoring the storage levels of general biomass types within the biogas plant and to analyze the fulfillment of TS-rate requirements within the wet process used in the biogas production (Hadin & Eriksson, 2016), the implementation of the depot class (Niemitalo & Ekkerman, 2022) was significantly expanded. The biogas plant is modeled as an object constructed by the depot class in the simulation. Methods for receiving biomass into storage and consuming it from storage were implemented, along with methods to track and update storage levels within the biogas plant and its TS rate. The necessary object variables were also defined (Eloranta, 2023b). These variables include the storage levels of the biogas plant, cumulative amounts of received biomasses, weighted average of the TS rate within the storage, cumulative dilution water consumed, and consumption rates. The consumption rates were defined through discussions with experts from LuKe and HAMK Bio, replicating the potential mixture utilization within the biogas reactor. In the simulation, the storage levels of biomasses decrease each day by the consumption rates, which represent the utilization of resources in the biogas production. When a vehicle arrives at the biogas plant to unload, the plant's biomass-receiving method is invoked. This action increases storage levels by the amount unloaded by the vehicle and updates the weighted average of the TS rate of the storage. The logging of storage levels and TS rate has been implemented to facilitate post-analysis for assessing the fulfillment of the 15% TS rate requirement. Additional object variables were introduced for the depot class to enable the identification of unfavorable actions within the biogas production process, similar to the cost variables in the optimizer's cost function discussed in more detail in the following subchapter 3.5.2. A cumulative amount of received biomass was implemented to track the moment when the received quantity of biomass reaches the yearly targeted input value. If this occurs, subsequent unloads executed by vehicles are unnecessary and are counted and penalized within the cost function. Overfilling of the biogas plant's storage is also considered as a cost variable within the cost function and is penalized on a daily basis. If the storage level of the biogas plant reaches zero, it causes a production stoppage, which is also penalized within the cost function based on the number of days with biogas production 57 stopped. To maximize the likelihood of generating an optimized routing with a continuously maintained TS rate below 15% within the storage, the cumulative consumption of dilution water in the biogas production has been introduced. Each day, if the storage's TS rate exceeds 15%, the total consumption of dilution water increases by the amount required to dilute the storage's biomass mixture to achieve a TS rate of 15%. The cumulative consumption of dilution water is incorporated as a single cost term within the cost function minimized by the GA, a topic that will be discussed in the next subsection. 3.5.2 Genetic algorithm in optimization A genetic algorithm (GA) implemented by Niemitalo & Ekkerman (2022) is utilized as part of this study to optimize the collection and logistic routing of biomasses in a way that minimizes the cost function to be defined. GAs are population-based metaheuristics approaches to solve optimization problems, suitable for variety of problems in the field of operations management (Katoch et al. 2021). During the optimization process of population-based metaheuristics, multiple potential solutions are simultaneously maintained, evaluated, compared, and evolved. The diversity is maintained in the population to avoid the solutions from being stuck in local optimum. The principles that guide the optimization of the genetic algorithm are derived from the biological evolution process. (Katoch et al. 2021) In short, GA’s optimization starts with a set of solutions, noted as the initial population. A single solution within the population is noted as a chromosome. By utilizing a set of genetic operators, new potential solutions, noted as children, are generated by combining the parts, noted as genes, of the initial solutions, noted as parents. The feasibility of a solution is examined with the defined fitness function. Based on the value of the fitness function, a solution is either accepted to the population or rejected. If the solution is accepted, a new generation of solutions is established. The process of inheriting potential solutions is repeated until the defined stopping criteria are met. (Gracia et al. 2014) The implementation of the GA includes a set of discrete phases. These phases encompass setting the parameters, the initialization of the population, i.e. selecting the initial parents, defining the fitness value, i.e. cost function based on the conditions and goals of the optimization context, defining the crossover and mutation operators utilized to inherit new solutions, and the replacement of each generation. (Gracia et al. 2014) The crossover and mutation operators are referred to as genetic operators, implemented not only to inherit descendants, i.e. generate new solutions, but also to maintain the diversity of the population and avoid ending up with a solution located in a local optimum (Gracia et al. 2014; Katoch et al. 2021). 58 The implementation of the GA utilized in this study is as follows. The parameters such as the population size, number of genes, and generations (Gracia et al. 2014) are set as follows: the number of genes, i.e. the number of potential locations to be visited by a vehicle (including pickup sites and the biogas plant), is calculated from the simulated logistic routing proposal. The size of the population is derived as follows: 𝑆𝑝= max (100,4𝑁𝑔) , (6) where 𝑆𝑝 is the size of the population used within the optimization, and 𝑁𝑔 is the number of genes calculated from the routing information. Based on the routing input information given to the algorithm, size of population within this study was calculated to be 12,384. The number of generations to be inherited serves as the stopping criteria for the optimization and was set to 600,000. After 400,000 generations were completed, the algorithm switched to a greedy approach for the remaining 200,000 generations, utilizing a greedy optimization algorithm. The difference between the non-greedy and greedy algorithms will be explained in the next paragraph. The initial population is generated randomly. For each initial solution, a random permutation of integers is generated from 0 to 𝑁𝑔−1, in which each integer represents one gene in the chromosome. Before inheriting descendants, i.e., utilizing the crossover operator, a selection of chromosomes from which the new solutions are generated is done. Selection is a crucial step in optimization if the GA is utilized since it determines which portion of the initial solutions will be included in the optimization process, and later on, the selection is applied with inherited generations. (Katoch et al. 2021) The selection procedure implemented by Niemitalo and Ekkerman (2022) and utilized within this study does not narrow the number of chromosomes while inheriting new generations of solutions. It takes each chromosome as a parent 𝑝0 and chooses another chromosome randomly as a parent 𝑝1, with each chromosome being chosen randomly only once. Therefore, each chromosome within the generation is utilized once as a parent 𝑝0 and once as a parent 𝑝1. However, after switching to a greedy approach, the algorithm fixes the choice of the other parent to be the best-known chromosome, i.e., the best among the simulated solutions thus far. (Niemitalo & Ekkerman, 2022) Consequently, the number of potential descendants equals the number of parents. By not narrowing the number of descendants within the inherited generations, the chances of ending up with a solution located in a local optimum are decreased since the best chromosome is chosen at the end of the optimization with the best fitness value, i.e., the lowest cost. The creation of descendants is done by the crossover operator after the selection procedure. The crossover operator determines the combination of genes for descendants, 59 consisting of parts of the parents' genes. (Gracia et al. 2014; Katoch et al. 2021) The crossover operator implemented by Niemitalo and Ekkerman (2022) and utilized within this study chooses two random crossover points from the chromosome of parent 𝑝1 and transmits the genes between the crossover points to the child. The rest of the genes are transmitted to the child from parent 𝑝0, from which the genes that are not in between the previously chosen crossover points are chosen. Thus, a new descendant is generated with genes consisting of parts of its parents' genes. Mutation operators could be implemented to add random variations to the genes of children with determined probabilities (Gracia et al. 2014). However, mutation operators were not implemented in the GA utilized within this study (Niemitalo & Ekkerman, 2022). Mutation operators would maintain the diversity within the population across the generations, ensuring the avoidance of ending up in a local optimum (Katoch et al. 2021). However, since the number of chromosomes within generations is not narrowed through the inheritance process, the implementation of a separate mutation operator was not seen as a necessity for this study. After inheriting descendants, the feasibility of each chromosome is evaluated using the cost function. The value of the cost function with the inherited chromosome is calculated and compared to the cost function values of its parents. The chromosome with the lowest cost function value is accepted into the new generation of solutions. (Niemitalo & Ekkerman, 2022) Therefore, it is ensured that each round of generations improves or maintains a set of solutions equally good as the previous generation by rejecting the child from the inherited generation if its cost is higher than either of its parents. The cost function utilized within this study is based on the one implemented by Niemitalo and Ekkerman (2022) and has been further developed for this study to better align its functioning with the context of transporting agricultural wastes and side-flows as biomasses for utilization in the biogas production. The cost function utilized within the optimization of this study is defined as follows: TC= 𝑐𝑜𝑣𝑙𝑑 ∑OVLDi 𝑚 𝑖=1 +Cvehicles + Cbgp, (7) where TC is the total costs caused by the system with the given routing, 𝑐𝑜𝑣𝑙𝑑 is the assumed cost of one day of overload at a pickup site, OVLDi is the number of days of overloading at pickup site 𝑖 and 𝑚 is the number of pickup sites. Cvehicles consist of costs caused by vehicles, and is defined as follows: 60 Cvehicles = 𝑝𝑓𝑐𝑓∑ODO𝑖 𝑛 𝑖=1 +𝑐𝑜𝑣𝑡ℎ ∑OVT𝑖 𝑛 𝑖=1 +𝑐𝑤𝑣 ∑WVi 𝑛 𝑖=1 , (8) where 𝑝𝑓 is the assumed price of fuel (€/l), 𝑐𝑓 is the assumed fuel consumption (l/100km), 𝑛 is the number of vehicles collecting and transporting the biomasses from farms to the biogas plant, ODO𝑖 is the total mileage driven by vehicle 𝑖 with the given routing, 𝑐𝑜𝑣𝑡ℎ is the assumed cost of the overtime work (€/h), OVT𝑖 is the total overtime work of vehicle 𝑖 with the given routing, 𝑐𝑤𝑣 represents the unit cost for a vehicle visiting the wrong type of pickup site, and WVi signifies the number of wrong visits carried out by vehicle 𝑖 based on the given routing. The incorporation of cost terms penalizing wrong visits was a component of the ongoing development of the cost function, which was undertaken as part of this study. Other cost terms associated with vehicles were introduced by Niemitalo & Ekkerman (2022), along with the cost term for pickup site overload, as presented in Equation 7. Cbgp was a part of further development of the cost function implemented by Niemitalo and Ekkerman (2022), and it represents costs caused by inefficiencies in biogas production at the biogas plant. Cbgp is defined as follows: Cbgp =𝑐𝑝𝑑𝑠𝑡PDST+c𝑜𝑣𝑓OVF+c𝑒𝑥𝑖𝑚EXIM+𝑐𝑑𝑤DW, (9) where 𝑐𝑝𝑑𝑠𝑡 is the assumed cost of a production stoppage, i.e. biogas plant running out of feedstock storage, PDST is the number of production stoppages that occurred within the simulation with the given routing, c𝑜𝑣𝑓 is the assumed cost of overfilling the biogas plant’s storage, which is assumed to have a capacity equal to the targeted annual inputs (Tampio, 2023), OVF is the number of overstocking occurrences that happened with the given routing, c𝑒𝑥𝑖𝑚 is the assumed cost of one unnecessary import of biomasses to the biogas plant after the targeted annual input is reached, EXIM represents the number of unnecessary imports that occurred within the simulation with the given routing, 𝑐𝑑𝑤 represents the assumed cost for one ton of dilution water consumed, and DW signifies the total amount of dilution water, in tons, consumed at the biogas plant based on the given routing. It should be noted that Equations 7, 8 and 9 all include penalty terms that are difficult to measure, but practical and reasonable to utilize within optimization procedure to ensure the streamlining of the related phases in the biogas supply chain. These penalty terms are costs of overloading of pickup sites, costs of wrong visits, costs of production stoppages and overfills of the feedstock storage at the biogas plant, and costs of unnecessary imports of biomasses. 61 Although Equation 7 shows the general cost term for overloads at pickup sites, it should be noted that overfilling of grass and straw pickup sites was neglected within the optimization. Since it was assumed that grasses and straws are baled on the fields, a theoretical infinite storage capacity could be assumed. Therefore, the cost function term OVLD𝑖’s value was not increased during the simulation if storage level exceeded site’s capacity at pickup site consisting of grass or straw, resulting in neglection of overloads at grass or straw pickup sites. The cost function presented in the Equation 7 was utilized in this study because it considers the dynamic optimization of logistic routing, aims to minimize transportation costs (mileage and overtime work), and consequently, aims to reduce transportation emissions (by minimizing mileage and fuel consumption). It also aims to enhance the flow of biomasses within the biogas supply chain by taking into account biomass flows at the biogas plant and addressing potential inefficiencies that might arise, including the consumption of dilution water and occurrences of production stoppages. These factors are consistent with those presented in Figure 5 of Chapter 2.6, which illustrates the factors within the biogas supply chain that can be improved using the methods developed and utilized in this study. The methods also consider case-specific factors related to the CBE and the transportation of agricultural wastes and side-flows as biomasses to be utilized in the biogas production, based on the literature review presented and discussed in Chapter 2. 3.6 Validity and reliability of the methodology Throughout the research process, ensuring high-quality standards was of utmost importance. This involved acknowledging the factors that influence the validity and reliability of the research and considering their potential threats. Validity is concerned with the accuracy and integrity of the research process, ensuring that it effectively addresses the intended subjects and problems through appropriate data collection and analysis methods, leading to credible and dependable results (Saunders et al. 2019). The examination of validity involves assessing whether the utilized methodologies effectively measure the intended variables and phenomena. Moreover, it includes evaluating the soundness of the analysis of results and the conclusions drawn from them. Additionally, the generalizability of the study's findings and methodologies to a broader context should be examined. The content of the study should also be validated, with the correct and accurate application of terms related to the research subject. Furthermore, it should be considered whether the utilized methods sufficiently cover the phenomenon under study. Ultimately, the goal is to have results that can be explained based on the utilized variables rather than the research process itself. (Saunders et al. 2019) 62 The methodology successfully covered the intended topics. The dynamics of CBE and the collection and transportation of agricultural wastes and side-flows as biomasses were analyzed comprehensively in Chapter 2's literature review. Insights derived from this review were then considered in later simulation schemes and optimization procedures, ensuring alignment of the upcoming results with the case-specific conditions and requirements. Ultimately, the key insights from the literature review are presented in Figure 5. By comparing it with the data processing methodologies in Chapter 3.4 used to locate biomass potentials within the relevant area, as well as the simulation schemes and GA's cost function in chapters 3.5, 3.5.1, and 3.5.2, all with case-specific adjustments, it can be argued that this study successfully aimed to measure and improve the aspects presented in Figure 5. However, it should be noted that this study has its restrictions regarding the quality of the initial geospatial data processed to model the pickup sites via clustering, as well as regarding the generalizability of the methods and the results, caused by implemented and assumed case-specific adjustments. The limitations of the study are discussed in more detail in Chapter 6.2. The replication and consistency of the study are referred to as reliability (Saunders et al. 2019). Due to the utilization of publicly available geospatial datasets for biomasses (LuKe, 2023), the development and utilization of quantitative methods for data procession with open-source code (Eloranta, 2023a), and the simulation approach to optimization utilizing a GA with open-source code (Niemitalo & Ekkerman, 2022; Eloranta, 2023b), replication of the study can be argued to be high. To replicate the study, one should possess basic knowledge of installing Python and C++ and related packages and running source code files utilizing those with the integrated development environment of their choice. Consistency of the study can also be argued to be high since the research process is described comprehensively and transparently step by step within this study. It starts with the literature review in Chapter 2 and forms the theoretical framework presented in Figure 5. It then continues by describing the optimization problem at a conceptual level and the data processing phases in detail in Chapters 3.2, 3.3, and 3.4, concluding with the model development by presenting the simulation approach and the GA to be utilized within the optimization in Chapters 3.5, 3.5.1, and 3.5.2. 63 4 RESULTS This chapter reports the results of the study. Initially, the results addressing the second and third research questions are presented in subsections 4.1 and 4.2. These results include the performance indicator values of the biogas supply chain, for which logistic routing was optimized. As the results are related to logistic routing planning on an annual basis, this chapter presents the selected performance indicators of the system produced with optimized routings. Comprehensive routing plans are excluded from the study as they exist in the form of raw data, including routes for nine vehicles each day within a year. The chapter closes by benchmarking the results with randomized and practical base case routings. 4.1 Optimized routing The optimized routing for collecting and transporting biomasses from sites to the biogas plant, without assuming the time-criticality of biomass, was addressed first, in response to the second research question. This optimized routing plan covers nine vehicles designated for collecting biomasses from sites, with an equal distribution among various biomass types. In other words, every third vehicle is assigned to collect grasses and straws, among others. The routing was optimized and presented using location indexes, which are employed in simulation and optimization, with each index representing a site generated within Chapter 3.4 or the biogas plant. The optimization trajectory is shown in Figure 16. This trajectory represents the lowest cost within the population of routing proposals, as a function of the number of cost function evaluations (Niemitalo & Ekkerman, 2022). The number of cost function evaluations increases by the population size of 12,384 within each generation during the optimization process. The number of generations was set to 600,000, serving as a stopping criterion for optimization. The switch to the greedy optimization algorithm can be observed in Figure 16 after 400,000 generations, resulting in an acceleration of the total cost minimization, which corresponds to 4,953,600 thousand cost function calculations. 70 Table 3. The vehicle-specific performance indicators under the assumption of biomass time-criticality. Type of the vehicle Mileage (km) Wrong visits Overtime (h) Mileage with the filtered routing (km) Vehicle 1 Slurry manure 927.42 14 0 329.45 Vehicle 2 Grass/straw 352.08 7 0 65.01 Vehicle 3 Dry manure 3,568.97 17 0 2,774.14 Vehicle 4 Slurry manure 1,124.62 14 0 235.92 Vehicle 5 Dry manure 4,599.48 10 0 4,132.51 Vehicle 6 Grass/straw 352.00 4 0 182.47 Vehicle 7 Slurry manure 559.80 14 0 148.62 Vehicle 8 Grass/straw 230.10 6 0 58.44 Vehicle 9 Dry manure 3,621.91 8 0 3,270.08 Table 4. The performance indicators of the biogas plant and vehicles under the assumption of biomass time-criticality. Optimized routing Optimized and filtered routing Total mileage (km) 15,336.4 11,196.65 Total wrong visits 94 0 Total overtime (h) 0 0 Number of days with no routes 74 95 Total pickup site overload days 10,666 10,666 Production stoppages 0 0 Consumption of dilution water (tons) 950,863 950,863 Unnecessary imports to the biogas plant 0 0 Overfillings within the biogas plant 0 0 Total costs (€) 5,381,630.34 5,287,626.20 Filtering out wrong visits improves the solution in terms of total costs, mileage, and the number of wrong visits. However, the results of the filtered routing should not be considered as the global optimum, as there are likely solutions with lower total costs. This is indicated by the high consumption of dilution water and the number of pickup site overload days in Table 4. Still, the goodness of the solution can be argued again with the number of production stoppages, total overtime hours of zero, and unnecessary imports to the biogas plant being zero. This implicates that with the given routing, continuous biomass flow and bi- 71 ogas production process was ensured within the biogas plant, simultaneously not exceeding the yearly targeted input values for biomasses at the biogas plant and not including routes exceeding the work shifts. 4.3 Benchmarking of results The results for both cases, as presented in Chapters 4.1 and 4.2, were validated and benchmarked by comparing the performance of optimized routings to randomized routings and routings that correspond to a practical approach to biomass routing, serving as a base case in benchmarking. In the randomized routing, the vehicle routing is chaotic. In the baseline routing, controlled collection of biomasses in response to orders arising randomly at the farms is assumed. These benchmarks were based on calculating performance indicator values, similar to the approach used in Chapters 4.1 and 4.2 with optimized routings. The benchmark routings used in these comparisons were generated by utilizing the same simulation and optimization model as presented in Chapters 3.5, 3.5.1, and 3.5.2. However, the number of generations to be optimized was set at 100, serving as a stopping criterion for the optimization process. Since the initial population is generated randomly by the GA (Gracia et al. 2014; Katoch et al. 2021), the solution obtained by stopping the optimization after 100 generations is also close to a random solution and is used as the randomized routing in benchmarks. Filtering out wrong visits from the randomized routing can be intuitively considered as the baseline solution for routing the collection of biomasses used in biogas production. In this scenario, orders for collecting biomasses from farms are generated randomly. Biomass pickup trucks are used to collect and transport the biomasses back to the biogas plant based on the orders. If there are multiple orders for the same truck on the same day, the truck collects them all within its capacity limits before returning to the biogas plant. Hence, the baseline routing can be considered a practical solution used by realworld logistical operators who collect biomasses in an order-controlled manner. 4.3.1 Benchmarks of the optimized routing Benchmarks for the optimized routing results presented in the Chapter 4.1, are shown below in Tables 5 and 6. Tables covers the performance indicators of the randomized and the baseline routing. 72 Table 5. The vehicle-specific performance indicators with the randomized routing (1) and baseline routing (2). Type of the vehicle Mileage (1) (km) Wrong visits (1) Overtime (1) (h) Mileage (2) (km) Overtime (2) (h) Vehicle 1 Slurry manure 3,841.94 76 0 74.26 0 Vehicle 2 Grass/straw 4,782.23 58 0 2,472.88 0 Vehicle 3 Dry manure 4,639.28 61 0 2,063.67 0 Vehicle 4 Slurry manure 4,210.87 85 0 15.56 0 Vehicle 5 Dry manure 4,796.84 62 0 1,833.60 0 Vehicle 6 Grass/straw 3,991.96 64 0 1,532.51 0 Vehicle 7 Slurry manure 4,732.29 96 0 77.98 0 Vehicle 8 Grass/straw 3,508.54 44 0 1,495.74 0 Vehicle 9 Dry manure 4,186.06 49 0 1,658.06 0 Table 6. The performance indicators of the biogas plant and vehicles with the randomized routing and baseline routing. Randomized routing Baseline routing Total mileage (km) 38,690.02 11,224.26 Total wrong visits 595 0 Total overtime (h) 0 0 Number of days with no routes 16 101 Total pickup site overload days 13,684 13,684 Production stoppages 53 53 Consumption of dilution water (tons) 7,252,490 7,252,490 Unnecessary imports to the biogas plant 0 0 Overfillings within the biogas plant 0 0 Total costs (€) 42,841,688.69 42,246,661.22 By comparing the performance of the optimized routings, as presented in Tables 1 and 2 in Chapter 4.1, with the performance of randomized and base case routings, as presented in Tables 5 and 6 above, the appropriateness of the developed simulation model and optimization method can be justified. In comparison to the practical baseline approach to biomass routing, where the collection of biomasses is triggered only by orders arising from the farms, a high number of production stoppages may occur within a year, as the baseline routing does not consider resource needs and production stoppages as a factor to be considered in routing. Such significant addition in the number of production 73 stoppages greatly contributes to the total costs of the system. Randomized routing generates even greater total costs since a high number of additional kilometers are driven by vehicles due to wrong visits. 4.3.2 Benchmarks of the optimized routing under the assumption of biomass time-criticality Benchmarks for the results that considered the time-criticality of biomass, as presented in Chapter 4.2, are shown below in Tables 7 and 8. Tables benchmarks the performance of the optimized results with the performances of the randomized and the baseline routing. Table 7. The vehicle-specific performance indicators with the randomized routing (1) and baseline routing (2) under the assumption of biomass time-criticality. Type of the vehicle Mileage (1) (km) Wrong visits (1) Overtime (1) (h) Mileage (2) (km) Overtime (2) (h) Vehicle 1 Slurry manure 4,103.60 86 0 123.99 0 Vehicle 2 Grass/straw 3,394.59 36 0 1,920.22 0 Vehicle 3 Dry manure 4,149.73 48 0 2,321.36 0 Vehicle 4 Slurry manure 3,583.74 75 0 59.69 0 Vehicle 5 Dry manure 4,486.19 44 0 2,371.47 0 Vehicle 6 Grass/straw 3,781.46 42 0 1,943.40 0 Vehicle 7 Slurry manure 4,688.62 99 0 124.90 0 Vehicle 8 Grass/straw 5,743.46 61 0 3,216.70 0 Vehicle 9 Dry manure 4,270.88 54 0 1,570.06 0 Table 8. The performance indicators of the biogas plant and vehicles with the randomized routing and baseline routing. Randomized routing Baseline routing Total mileage (km) 38,202.26 13,651.79 Total wrong visits 545 0 Total overtime (h) 0 0 Number of days with no routes 14 88 Total pickup site overload days 13,018 13,018 Production stoppages 34 34 Consumption of dilution water (tons) 7,995,460 7,995,460 Unnecessary imports to the biogas plant 0 0 Overfillings within the biogas plant 0 0 Total costs (€) 44,573,238.2 44,028,213.65 74 By comparing the performance of the optimized routings under the assumption of biomass time-criticality, as presented in Tables 3 and 4 in Chapter 4.2, with the performance of the randomized and the baseline routings under the same assumption of biomass time-criticality, as presented in Tables 7 and 8 above, similar conclusions can be drawn regarding the appropriateness of the developed simulation and optimization model. If the collection of biomasses is triggered solely by orders from farms occurring randomly, a high number of production stoppages may occur within a year, significantly contributing to the total costs of the system. Randomized routing introduces additional costs due to the extra kilometers driven during wrong visits. 75 5 DISCUSSION This section analyzes and compares the results of the study presented in previous chapter. Finally, the chapter presents and discusses the key findings of the study. 5.1 Comparison of results By comparing the results presented in Chapters 4.1 and 4.2, it is possible to develop perceptions of the impact on solutions, which may be influenced by the consideration of biomass time-criticality. Interestingly, the total distance driven by the vehicles and the number of visits to wrong types of sites by vehicles appear to be significantly lower when assuming biomass time-criticality. However, it remains unclear whether the difference between mileage and wrong visits is solely due to the assumption of biomass time-criticality. The assumption of biomass time-criticality ultimately incorporates the continuous drying process of biomass within the sites, vehicles, and the biogas plant's storage into the simulation model, as presented in Equations 3 and 4, inspired by the biomass storage model by van Dyken et al. (2010). This process leads to a continuous decrease in the volume of biomass and an increase in its TS rate due to drying. Thus, it cannot be explicitly concluded that there is a cause-and-effect relationship between the drying process of biomasses and the mileage driven by vehicles transporting them based on these results. The possible improvement in mileage and wrong incorrect visits may also be attributed to pure coincidence, as there is a chance for the GA to become stuck in local optima if the optimization is terminated prematurely (Katoch et al. 2021). Consequently, further research on this topic is necessary. It can be assumed that in a time-critical scenario, biogas yield should increase when the effects of time-criticality are minimized through efficient logistic routing. Therefore, if the increase in biogas yield exceeds the additional logistic costs incurred by driving extra kilometers, then it would be justifiable to do so. However, it's important to note that the model used in this study does not measure the actual amount of biogas produced within the plant; it only tracks the consumption of biomasses as resources. Consequently, conclusions about this observation cannot be drawn based on the study's results. The need for further research on the topic is justified once again. However, based on these results, it can be stated that the passive processes occurring within the biomasses should be considered when planning the logistic routing of a circular system for transporting biomasses to biogas production. In both cases, the number 76 of production stoppages occurring within a year was managed to be reduced to zero with the optimized routing. Additionally, the consumption of dilution water and the total system costs appear to be lower when optimizing the routing under the assumption of biomass time-criticality. The actual availability of biomasses for use in biogas production may also be influenced by time-critical factors, as suggested by the lower number of overload days at the pickup sites when biomass time-criticality was considered. This can be explained by the assumption of biomass time-criticality. Storage levels at the pickup sites continuously decrease daily, resulting in fewer days when storage levels exceed their capacities. Given the relatively high number of days with pickup site overloads in both cases, one might question whether nine vehicles are enough to collect biomasses from this many sites. On the other hand, the number of pickup sites was intentionally set to be this high to ensure an adequate supply of biomass for the annual resource consumption at the biogas plant. Therefore, using the number of pickup site overload days as a performance indicator and cost factor in this case may not be necessary. Based on this, its high value with the optimized routing can be partially ignored. While part of the reduced cost is attributed to lower mileage and fewer wrong visits when considering biomass time-criticality, the key observation here is that the number of production stoppages remains at zero in both cases. This indicates a continuous biogas production process and biomass flow within the biogas plant. This highlights the potential cost savings through more efficient logistic planning, as it becomes evident that passive time-critical processes affecting the quality of biomasses for energy production do exist (van Dyken et al. 2010; Mönch-Tegeder et al. 2013; Hadin et al. 2016). The question, then, is whether decision-makers planning the logistic systems used within circular supply chains for the biogas production take these processes into consideration or not. Additionally, after filtering out wrong visits from the optimized routings and rerunning the simulation with the filtered routings, the results appear to be promisingly good. It should be mentioned that wrong visits should be considered more as a feature of a solution that exists due to the deficiency of the GA. The GA utilized within the study’s optimization generates routing proposals solely based on the locations that are given to it as an input. Therefore, it does not check or exclude the possible wrong visits arising within the generated routing, which is why the filtering procedure executed afterwards is appropriate. Wrong visits arising within the optimized routings indicates the suboptimality of the solution, since if the global optimum was found by the GA, wrong visits should not occur at all. 77 In real life, logistics operators know which type of sites they should visit, and therefore, wrong visits would not occur. This is also why filtering out wrong visits from the optimized solutions is justified. Continuous biogas production is ensured in both cases with no need for overtime work, and vehicles visit sites only of their type, reducing the kilometers and, therefore, total costs significantly. It remains unclear how far these results with the filtered routings are from the global optimum, as there is no reason to believe they are at the global optimum. On the other hand, it's not clear what cost factors the solutions could be improved upon. The need for further research on this topic is justified, as the development of methods for finding the global optimum solution is required. The benchmarks of the results prove the benefits of planning the logistic routing of biomasses by utilizing dynamic optimization methods. As can be concluded from the benchmarks presented in Chapter 4.3, if the decision-making process for collecting biomasses is simply reactive to orders arising randomly from the farms, the continuous process of biogas production may be endangered, as observed with the high number of production stoppages within the benchmarks. In addition, excess kilometers are driven due to the reactive approach of collecting biomasses in an order-controlled manner, instead of planning the logistic routing while considering the resource needs and consumption within the biogas plant. Additionally, if the routing is completely random, a significant number of additional kilometers may be driven due to the wrong visits, decreasing the performance of the solution. On the other hand, the optimized routings are generated proactively by considering resource shortages, consumption of dilution water, total mileage driven by vehicles, and other relevant factors. The decision-making process for collecting the biomasses is more rational and sophisticated in comparison to reactive collection in an order-driven manner, which may significantly decrease the total costs of the system, ultimately making the circular supply chain more effective and enhancing the practices of the CBE. 5.2 Key findings The objectives of the study included identifying critical general and case-specific factors for optimizing the logistic routing within a circular supply chain to ensure its compatibility with the CBE practices. This objective partially overlapped with the subobjectives of implementing spatial information mapping for biomasses located near the biogas plant, generating information about biomass availability, and subsequently identifying potential strategic partnerships for the case company. Using spatial information mapping of biomasses, the study aimed to plan and optimize logistic routing crucial to the biogas plant's production process. This optimization sought to achieve a continuous biogas production 78 and efficient biomass flow while simultaneously minimizing logistic costs by planning and optimizing routes for vehicles collecting and transporting biomasses from various sites to the biogas plant. Additionally, the study aimed to advance understanding of how passive time-critical processes occurring within the biomasses influence the optimization of logistic routing. The objectives of the study were relatively well achieved. The factors needed to align logistic system operations with CBE practices were effectively addressed. This resulted in no significant deficiencies in the developed methodology, which models, simulates, and optimizes the functions of the components of the circular supply chain responsible for transporting biomass from farms to the biogas plant for biogas production. For example, the requirement that the TS-rate must be at most 15% within the biogas production process (Hadin & Eriksson, 2016) was probably the most important restriction related to CBE practices that was identified and considered in this study. Other identified and considered factors influencing the implementation of the CBE practices are presented in Figure 5 in Chapter 2.6. The implementation of spatial information mapping for biomasses near the case company’s biogas plant was successfully completed. As presented in Chapter 3.4, the phases during the data processing achieved the goal of mapping biomass potentials near the biogas plant. This process began with wide and unmapped datasets of biomasses located in Finland and concluded with mapping the biomass potentials within a 50 km radius of the biogas plant. It also proposed locations from which the biomasses should be picked up to meet the annual demand for biomasses within the biogas production process. Similar maps as presented in Chapter 3.4, can be found for all inspected biomasses of the study in GitHub (Eloranta, 2023c). The study optimized and proposed a routing plan for the case company. By following this plan, logistic routing could be executed relatively cost-effectively, ensuring continuous biogas production at the biogas plant. The performance of the optimized routing plans was validated and assessed by comparing the values of selected performance indicators with randomized routings and routings corresponding to a practical approach to biomass routing. These performance indicators were the cost terms implemented within the simulation model, which were minimized during the routing optimization process. However, as demonstrated in Chapter 4, the optimized routings, which serves as the results of this study, may represent suboptimal solutions to the problem of BRBPP. Thus, the results likely provide sufficiently decent solutions for executing logistic operations, but they probably may be further optimized with additional computing time and power. Still, the results of the study successfully enable improving the efficiency of the case’s circular supply 79 chain, as the optimized routing improve the factors presented in Figure 5 in Chapter 2.6 simultaneously considering the relevant factors of implementation of the CBE practices. The final objective of the study was to examine how the consideration of time-critical factors, occurring as passive drying processes within the biomasses collected, transported, and used in biogas production, affects the solution of optimized routing. Based on the study's results, it may be concluded that considering passive processes that affect biomass quality in terms of their bioenergy potential could potentially lead to cost savings through more efficient logistic routing planning. However, since the study's results cannot be considered as the global optimum, a direct causal link between considering timecritical factors and reduced total system costs cannot be definitively established. Therefore, further research on this topic is required to authenticate the global optimality of the optimized routing. 86 Giampietro, M. 2019, “On the Circular Bioeconomy and Decoupling: Implications for Sustainable Growth”, Ecological Economics, vol. 162, pp. 143–156. Government Decree on the Restriction of Discharge of Nitrates From Agriculture into Waters 2000/931. Issued in Helsinki 9.11.2000. Available at: https://www.finlex.fi/en/laki/kaannokset/2000/en20000931?search%5Btype%5D=pika&search%5Bkieli%5D%5B0%5D= en&search%5Bpika%5D=931%2F2000. (Accessed at: 25.7.2023) Gracia, C., Velázquez-Martí, B. & Estornell, J. 2014, “An application of the vehicle routing problem to biomass transportation”, Biosystems Engineering, vol. 124, pp. 40–52. Grafström, J. & Aasma, S. 2021, “Breaking circular economy barriers”, Journal of Cleaner Production, 292126002. Hadin, Å. & Eriksson, O. 2016, ”Horse manure as feedstock for anaerobic digestion”, Waste Management, vol. 56, pp. 506–518. Hadin, Å., Eriksson, O. & Hillman, K. 2016, ”A review of potential critical factors in horse keeping for anaerobic digestion of horse manure” Renewable and Sustainable Energy Reviews, vol. 65, pp. 432–442. Hojnik, J. & Ruzzier, M. 2016, “What drives eco-innovation? A review of an emerging literature”, Environmental Innovation and Societal Transitions, vol. 19, pp. 31–41. Huang, Y., Fan, Y. & Chen C. 2014, “An Integrated Biofuel Supply Chain to Cope With Feedstock Seasonality and Uncertainty”, Transportation Science, vol. 48, No. 4, November 2014, pp. 540–554. Jeffares, A. 2019, “K-means: A Complete Introduction”, Towards Data Science. Available at: https://towardsdatascience.com/k-means-a-complete-introduction-1702af9cd8c. (Accessed at: 9.6.2023) Jensen, I., Münster, M. & Pisinger, D. 2017, ”Optimizing the supply chain of biomass and biogas for a single plant considering mass and energy losses”, European Journal of Operational Research, vol. 262, pp. 744–758. Jozefowiez, N., Semet, F. & Talbi E. 2008, ”Multi-objective vehicle routing problems”, European Journal of Operational Research, vol. 189, pp. 293–309. Jumppanen, A. 2023, “Vaihtoehto viljalle: Golden Fields Factory tavoittelee 20 000 hehtaarin nurmialaa”, AGRImedia. Available at: https://www.agrimedia.fi/golden-fields-factory-tavoittelee-20-000-hehtaarin-nurmialaa/. (Accessed 23.5.2023) 87 Kamilaris, A. & Prenafeta-Boldú, F. X. 2021, ”Examining the perspectives of using manure from livestock farms as fertilizer to crop fields based on a realistic simulation”, Computers and Electronics in Agriculture, 191106486. Karmakar, S., Kumar De, S. & Goswami A. 2017, ”A pollution sensitive dense fuzzy economic production quantity model with cycle time dependent production rate”, Journal of Cleaner Production, vol. 154, pp. 139–150. Katoch, S., Chauhan, S. & Kumar, V. 2021 ”A review on genetic algorithm: past, present, and future”, Multimedia Tools and Applications, vol. 80, pp. 8091–8126. Kinyua, M., Wald, I., Camacho-Céspedes, F., Izurieta, R., Haas, C. & Ergas, S. 2016, "Does the use of tubular digesters to treat livestock waste lower the risk of infection from Cryptosporidium parvum and Giardia lamblia?", Journal of Water and Health, vol. 14 (5), pp. 738–753. Kirchherr, J., Piscicelli, L., Bour, R., Kostense-Smit, E., Muller, J., Huibrechtse-Truijens, A. & Hekkert, M. 2018, “Barriers to the Circular Economy: Evidence From the European Union (EU)”, Ecological Economics, vol. 150, pp. 264–272. Korhonen, J., Honkasalo, A. & Seppälä, J. 2018, ”Circular Economy: The Concept and its Limitations”, Ecological Economics, vol. 143, pp. 37–46. Kymäläinen, M. & Tampio, E. 2023, ”BioKanta – Biokaasua hiiliviisaasti ja ravinteet kiertoon Kanta-Hämeessä”. Häme University of Applied Sciences (HAMK). Available at: https://www.hamk.fi/biokanta. (Accessed 5.4.2023) Lessman, M., Kanellopoulos, J., Kros, J., Orsi, F. & Bakker, M. 2023, “Maximizing agricultural reuse of recycled nutrients: A spatially explicit assessment of environmental consequences and costs”, Journal of Environmental Management, 332117378. Liu, S., Zhang, Y., Liu, Y., Wang, L., Wang, X. 2019, “An ‘Internet of Things’ enabled dynamic optimization method for smart vehicles and logistics tasks”, Journal of Cleaner Production, vol. 215, pp. 806–820. Liu, Y., Gilchrist, A., Zhang, J. & Li, X. 2008, ”Detection of Viable but Nonculturable Escherichia coli O157:H7 Bacteria in Drinking Water and River Water”, Applied and Environmental Microbiology, vol. 74 (5), pp. 1502–1507. LuKe, 2023, Biomassa-atlas. Available at: https://biomassa-atlas.luke.fi/. (Accessed 9.6.2023) 88 Luostarinen, S., Grönroos, J., Hellstedt, M., Nousiainen, J. & Munther, J. 2017, “Finnish Normative Manure System: System documentation and first results”, Natural resources and bioeconomy studies, vol. 48/2017. Natural Resources Institute Finland. Helsinki. Magiya, J. 2019. “Clustering GPS Coordinates and Forming Regions with Python”, Level Up Coding by gitconnected.com. Available at: https://levelup.gitconnected.com/clustering-gps-co-ordinates-forming-regions-4f50caa7e4a1. (Accessed at: 9.6.2023) Manyi-Loh, C., Mamphweli, S., Meyer, E., Okoh, A., Makaka, G. & Simon, M. 2013, “Microbial Anaerobic Digestion (Bio-Digesters) as an Approach to the Decontamination of Animal Wastes in Pollution Control and the Generation of Renewable Energy”, International Journal of Environmental Research and Public Health, vol. 10, pp. 4390–4417. Mohan, S., Chiranjeevi, P., Dahiya, S. & Kumar, A. 2018, “Waste derived bioeconomy in India: A perspective”, New Biotechnology, vol. 40, pp. 60–69. Monlau, F., Sambusiti, C., Ficara, E., Aboulkas, A., Barakat, A. & Carrère. 2015, ”New opportunities for agricultural digestate valorization: current situation and perspectives”, Energy & Environmental Science, vol. 8, pp. 2600–2621. Monteiro, E. & Ferreira, S. 2022, “Biomass Waste for Energy Production”, Energies, vol. 15 (16), pp. 5943. Mönch-Tegeder, M., Lemmer, A., Oechsner, H. & Jungbluth, T. 2013, ”Investigation of the methane potential of horse manure”, Agricultural Engineering International, vol. 15, n. 2, pp. 161–172. Nag, R., Auer, A., Markey, B., Whyte, P., Nolan, S., O’Flaherty, V., Russell, L., Bolon, D., Fenton, O., Richards, K. & Cummins, E. 2019, “Anaerobic digestion of agricultural manure and biomass – Critical indicators of risk and knowledge gaps”, Science of the Total Environment, vol. 690, pp. 460–479. Niemitalo, O. & Ekkerman, G. 2022, “Logistics simulation and optimization”. Source code and the documentation of the original simulation and optimization models. Available at: https://github.com/hamk-uas/logistics_sim. (Accessed at: 20.7.2023) openrouteservice, 2023, “Routing API”. Available at: https://openrouteservice.org/. (Accessed at: 20.7.2023) Paredes-Sánchez, J., López-Ochoa, L., López-González, L., Las-Heras-Casas, J. & Xiberta-Bernat, J. 2019, “Evolution and perspectives of the bioenergy applications in Spain”, Journal of Cleaner Production, vol. 213, pp. 553–568. 89 Ponjavic, M., Karabegovic, A., Celebicanin, S. & Ljusa, M. 2020, “Performance indicators for poultry manure supply model analysis”, UPB Scientific Bulletin, Series D: Mechanical Engineering, vol. 82, iss. 4. Prieto-Sandoval, V., Jaca, C. & Ormazabal, M. 2018, “Towards a consensus on the circular economy”, Journal of Cleaner Production, vol. 179, pp. 605–615. Ranta, V., Aarikka-Stenroos, L. & Väisänen, J-M. 2021, ”Digital technologies catalyzing business model innovation for circular economy – Multiple case study, Resources, Conversation & Recycling, 164105155. Saji, B. 2023, “Elbow Method for Finding the Optimal Number of Clusters in K-Means” Analytics Vidhya. Available at: https://www.analyticsvidhya.com/blog/2021/01/in-depthintuition-of-k-means-clustering-algorithm-in-machine-learning/. (Accessed: 15.6.2023) Sakar, S., Yetilmezsoy, K. & Kocak, E. 2009, “Anaerobic digestion technology in poultry and livestock waste treatment – a literature review”, Waste Management & Research, vol. 27 (1), pp. 3–18. Salvador, R., Puglieri, F., Halog, A., de Andrade, F., Piekarski, C. & Francisco, A. 2021, ”Key aspects for designing business models for a circular bioeconomy”, Journal of Cleaner Production , vol. 278, 124341. Saunders, M., Lewis, P. & Thornhill, A. 2019, “Research methods for business students”. 8th edition. Harlow: Pearson. Schütz, F. 2021, “SimCpp20 – a discrete-event simulation framework for C++”. Available at: https://github.com/fschuetz04/simcpp20. (Accessed: 20.7.2023) scikit-learn developers, 2023, sklearn.cluster.KMeans, scikit-learn. Available at: https://scikit-learn.org/stable/modules/generated/sklearn.cluster.KMeans.html. (Accessed at: 9.6.2023) Sevgen, A. & Sargut, F. Z. 2019, ”May reorder point help under disruptions?”, International Journal of Production Economics, vol. 209, pp. 61–69. Shabani, N. & Sowlati, T. 2013, “A mixed integer non-linear programming model for tactical value chain optimization of a wood biomass power plant”, Applied Energy, vol. 104, pp. 353–361. Stahel, W. R. 2016, “The circular economy”, Nature, vol. 531 (7595), pp. 435–438. Stevenson, W. J. 2014, “Operations management”, 12th ed., global ed. New York, NY: McGraw Hill/Irwin. 90 Su, G., Ong, H., Zulkifili, N., Ibrahim, S., Chen, W., Chong, C. & Ok, Y. 2022, “Valorization of animal manure via pyrolysis for bioenergy: A review”, Journal of Cleaner Production, vol. 343, 130965. Sun, P., Hu, Y., Lan, J., Tian, L. & Chen, M. 2019, ”TIDE: Time-relevant deep reinforcement learning for routing optimization”, Future Generation Computer Systems, vol. 99, pp. 401–409. Tampio, E. 2023, “Biokaasua hiiliviisaasti ja ravinteet kiertoon Kanta-Hämeessä – BioKanta – TP1.3 Humppilan casen aloituskokous 23.3.2023”. The presentation material for the BioKanta project meeting. Team SimPy, 2020, “Discrete event simulation for Python”. Available at: https://simpy.readthedocs.io/en/latest/. (Accessed: 20.7.2023) van Dyken, S., Bakken, B. & Skjelbred, H. 2010, ”Linear mixed-integer models for biomass supply chains with transport, storage and processing”, Energy, vol. 35, pp. 1338– 1350Varenova, D., Samy, M. & Combs, A. 2013, “Corporate social responsibility and profitability: trade-off or synergy - Perceptions of executives of FTSE All-Share companies”, Sustainability Accounting, Management and Policy Journal, vol. 4 (2), pp. 190–215. Velázquez-Marti B. & Annevelink, E. 2009, “GIS application to define biomass collection points as sources for linear programming of delivery networks”, Transactions of ASABE, vol. 52 (4), pp. 1069–1078. Yu, B., Liu, X., Ji, C. & Sun H. 2023, ”Greenhouse gas mitigation strategies and decision support for the utilization of agricultural waste systems: A case study of Jiangxi Province, China”, Energy, vol. 265, 126380. Yuan, Z., Bi, J. & Moriguichi, Y. 2006, “The Circular Economy: A New Development Strategy in China”, Journal of Industrial Ecology, vol. 10 (1-2), pp. 4–8. Wang, Z., Bui, Q., Zhang, B. & Pham, T. 2020, “Biomass energy production and its impact on the ecological footprint: An investigation of G7 countries”, Science of the Total Environment, 140741. Wei, X., Liu, Q., Pu, A., Wang, S., Chen, F., Zhang L., Zhang, Y., Dong, Z. & Wan X. 2022, “Knowledge Mapping of bioeconomy: A bibliometric analysis”, Journal of Cleaner Production, 373133824. 91 Zhang, J., Osmani, A., Awudu, I. & Gonela, V. 2013, ”An integrated optimization model for switchgrass-based bioethanol supply chain”, Applied Energy, vol. 102, pp. 1205– 1217.