Bayesian material flow analysis for systems with multiple levels of disaggregation and high dimensional data

Junyang Wang, Kolyan Ray, Pablo R. Brito‐Parada, Yves Plancherel, Tom Bide, J.m. Mankelow, John E. Morley, Julia A. Stegemann, Rupert J. Myers

Abstract

Material Flow Analysis (MFA) is used to quantify and understand the life cycles of materials from production to end of use, which enables environmental, social and economic impacts and interventions. MFA is challenging as available data is often limited and uncertain, leading to an underdetermined system with an infinite number of possible stocks and flows values. Bayesian statistics is an effective way to address these challenges by principally incorporating domain knowledge, and quantifying uncertainty in the data and providing probabilities associated with model solutions. This paper presents a novel MFA methodology under the Bayesian framework. By relaxing the mass balance constraints, we improve the computational scalability and reliability of the posterior samples compared to existing Bayesian MFA methods. We propose a mass based, child and parent process framework to model systems with disaggregated processes and flows. We show posterior predictive checks can be used to identify inconsistencies in the data and aid noise and hyperparameter selection. The proposed approach is demonstrated on case studies, including a global aluminium cycle with significant disaggregation, under weakly informative priors and significant data gaps to investigate the feasibility of Bayesian MFA. We illustrate just a weakly informative prior can greatly improve the performance of Bayesian methods, for both estimation accuracy and uncertainty quantification.

Introduction

Material Flow Analysis (MFA) is a vital tool for the circular economy, tracking how substances like aluminum or zinc move from extraction to manufacturing and eventual disposal. However, MFA faces a significant data problem: most systems are "underdetermined," meaning there are far more unknown flows than actual measurements. Traditional methods often struggle with data disaggregation—where data exists at varying levels of granularity—and computational instability in high-dimensional spaces.

Methodology: A Scalable Bayesian Framework

To address these challenges, the authors introduce a robust Bayesian framework that incorporates domain knowledge and quantifies uncertainty. The workflow, depicted in Figure 1, begins with prior parameters specified via domain knowledge. These priors are combined with observed data through a likelihood function to produce posterior distributions for all change in stock and flow variables using Bayes' Theorem.

The Bayesian MFA framework: from domain-specified prior parameters to posterior distributions.

A key innovation is the use of the No-U-Turn Sampler (NUTS), an advanced Hamiltonian Monte Carlo algorithm. Unlike older methods that struggle in high-dimensional spaces, NUTS utilizes gradient information to navigate hundreds of variables efficiently, allowing for scalable uncertainty quantification.

Handling Disaggregation and Mass Balance

The framework introduces two structural innovations to modernize MFA:

  1. Noisy Mass Balance: Instead of rigid mass-balance constraints (Inflows = Outflows), the model introduces a small noise term. This relaxation represents "epistemic uncertainty"—the reality that system diagrams are simplified models. This approach prevents the creation of "zero-probability" regions that hinder computational convergence.
  2. Child and Parent Framework: The system is treated as a hierarchy. By modeling at the most granular "child" level, aggregated "parent" data can be incorporated as the sum of its children. This allows the model to ingest data regardless of its hierarchical level.
A simplified process-flow diagram illustrating the conservation of mass within a system.

As shown in Figure 2, the conservation of mass for a specific process (e.g., Process 1) is defined by the relationship between its inflows ($U_{0,1}$), outflows ($U_{1,2}, U_{1,3}$), and stock changes ($S_1$).

Results and Insights

The methodology was validated using a global aluminum cycle dataset (approx. 180 variables) and a simulation of the China zinc cycle.

  • Computational Efficiency: The model successfully sampled the complex aluminum system in approximately 90 minutes on a standard laptop, a significant improvement over previous methods that often failed to converge on systems of this size.
  • The Power of Weakly Informative Priors: A striking finding was that simply knowing the order of magnitude of a flow (e.g., ~10 Mt vs ~100 Mt) is as valuable as 10-12 actual data points. This highlights the importance of domain knowledge in data-poor environments.
  • Error Detection: Posterior predictive checks identified discrepancies in reported data. For instance, in the aluminum case study, the model flagged an inconsistency where reported inflows for "Internal Remelting" did not match reported outflows, serving as a diagnostic tool for data quality.

Conclusion

This research demonstrates that Bayesian MFA, when implemented with relaxed mass-balance constraints and efficient sampling algorithms, provides a scalable and reliable solution for quantifying global material cycles. By moving away from rigid transfer coefficients to a mass-based parametrization, the authors avoid mathematical singularities and improve accuracy. Future work may involve automating prior specification through expert elicitation and extending the model to dynamic, time-series analysis.

Summary updated

Abstract

Material Flow Analysis (MFA) is used to quantify and understand the life cycles of materials from production to end of use, which enables environmental, social and economic impacts and interventions. MFA is challenging as available data is often limited and uncertain, leading to an underdetermined system with an infinite number of possible stocks and flows values. Bayesian statistics is an effective way to address these challenges by principally incorporating domain knowledge, and quantifying uncertainty in the data and providing probabilities associated with model solutions. This paper presents a novel MFA methodology under the Bayesian framework. By relaxing the mass balance constraints, we improve the computational scalability and reliability of the posterior samples compared to existing Bayesian MFA methods. We propose a mass based, child and parent process framework to model systems with disaggregated processes and flows. We show posterior predictive checks can be used to identify inconsistencies in the data and aid noise and hyperparameter selection. The proposed approach is demonstrated on case studies, including a global aluminium cycle with significant disaggregation, under weakly informative priors and significant data gaps to investigate the feasibility of Bayesian MFA. We illustrate just a weakly informative prior can greatly improve the performance of Bayesian methods, for both estimation accuracy and uncertainty quantification.
BibTeX
@article{wang2024bayesian,
  author = {Junyang Wang and Kolyan Ray and Pablo R. Brito‐Parada and Yves Plancherel and Tom Bide and J.M. Mankelow and John E. Morley and Julia A. Stegemann and Rupert J. Myers},
  title = {{Bayesian material flow analysis for systems with multiple levels of disaggregation and high dimensional data}},
  journal = {Journal of Industrial Ecology},
  year = {2024},
  doi = {10.1111/jiec.13550},
}
Figure