Modeling of sustainable materials management systems: Hybrid science-based, data-driven approaches

Fueled by growing waste generation, worsening impacts of climate change, and tightening resource constraints, societies are beginning to pivot to a circular economy, of which sustainable materials management (SMM) is a crucial component. Under this SMM paradigm, solid waste management systems (SWMS) need to serve the dual purposes of waste disposal and resource valorization in technologically feasible, economically beneficial, environmentally responsible, and socially benign ways. This requires innovative holistic approaches for understanding the flows and stocks of resources, deploying appropriate treatment and valorization technologies, and mobilizing social and financial capital. To advance the scientific understanding of SMM and to tackle the dilemma of coupling between complex science and scarce data, this dissertation leverages system dynamics perspectives and data-driven quantification capabilities to develop novel science-based, data-driven models for supporting real-world SMM decisions.As a prerequisite of SMM, a hybrid demand forecast framework is built by integrating empirical evidence-informed future scenarios into a stocks dynamics model to project future stocks and flows of materials. This adaptable framework is applied to forecast the U.S. copper demand and stocks from 2016 to 2070, which reveals a stabilizing trend of per capita in-use stock and growing trends of consumption and end-of-life (EOL) copper generation. By analyzing various U.S.-specific future scenarios, the recovery of EOL materials is identified as a main sustainability challenge. To understand the mechanisms of EOL recovery and the potential of data-driven modeling, a comprehensive global knowledge base is built by reviewing and compiling data on municipal solid waste (MSW) generation, characterization, and management practices. A total of 1,720 records from various reputable sources, including NGO databases, governmental reports, and peer-reviewed articles are collected to build statistical models that predict MSW generation, composition, and management practices using 18 socioeconomic variables. This knowledge base not only systemizes sub-domains of SWMS knowledge into a comprehensive framework, but also highlights pervasive data and knowledge gaps. To address the need for decision support in the new era of SMM while facing the challenge of insufficient data, a novel hybrid machine learning model is built by imposing the holistic decision-making context of SWMS on a traditional neural network architecture. The hybrid model is capable of learning various technical, economic, and behavioral aspects of SWMS using the relatively small and heterogeneous dataset in the knowledge base. In comparison, the hybrid model not only outperforms traditional machine learning models with similar complexity in prediction errors and convergence rates, but also offers superior interpretability for policy decision support.In summary, this dissertation is intended to empower SMM decision makers with a suite of hybrid tools for forecasting material flows and stocks, revealing the data landscape of EOL streams, and modeling the entire decision-making context of SWMS. From the methodology innovation standpoint, this dissertation pioneers and contributes to the nascent research field of integrating sustainability science with data science.