Building a Life Cycle Assessment Model using Bayesian Networks

This paper introduces the Oilfield Pollutant Graphical Model (OPGM), an innovative approach designed to improve the benchmarking and uncertainty analysis of greenhouse gas (GHG) emissions in oilfields. Building on the robust foundation provided by the Oil Production Greenhouse Gas Emission Estimator (OPGEE) framework, OPGM retains all essential functionalities of the latest OPGEE iteration (v3.0c), while offering substantial improvements in user experience and computational performance. The integration of OPGM into existing Life Cycle Assessment (LCA) practices holds the promise of significantly improving the precision and speed of environmental impact analyses, offering a vital tool for policymakers and industry stakeholders in their efforts to better understand and manage the environmental impacts of oilfield operations.

Introduction

Life Cycle Assessment (LCA) is a systematic approach to evaluating the environmental impacts of a product or process throughout its entire life cycle, from the extraction of raw materials to the disposal of waste. Oil fields are complex systems with a wide range of environmental impacts, including greenhouse gas emissions, water pollution, and land degradation. According to the International Energy Agency (IEA) [6], oil and gas production, transport and processing resulted in 5.1 billion tonnes (Gt) of equivalent CO2 (CO2eq.) in 2022 - just under 15% of global energy sector GHG emissions. LCA is a valuable tool for understanding and managing these impacts. We propose a bottom-up engineering-based emission modeling to evaluate the scope 1, scope 2, and scope 3 emissions of an oil operation from exploration to extraction development operations, surface handling and transport.Probabilistic Graphical Models (PGM) (also called Bayesian Belief Networks [1]) can be used to represent complex systems and make predictions about their behavior [7]. We present a new LCA model (OPGM) built using the PGM library TKRISK?. We show that the PGM can accurately capture the complexity of a well-established reference model while opening each of its components to the layman's eyes. OPGM presents a scalable alternative to existing solutions and can be used to perform advanced evaluations, including sensitivity analysis and uncertainty quantification.

Related work

The Oil Production Greenhouse Gas Emission Estimator (OPGEE) is an engineering-based life cycle assessment (LCA) tool that is primarily used to estimate upstream greenhouse gas (GHG) emissions. OPGEE is recognized for its bottom-up LCA approach and has gained widespread adoption across various sectors. Regulators, operators, consulting firms, financial institutions, autonomous intergovernmental organizations, and non-governmental organizations (NGOs) have implemented OPGEE in their operations [8]. Among its users are the California Air Resources Board (CARB) under the Low Carbon Fuel Standard (LCFS), Chevron, McKinsey & Company, S&P Global Platts, the International Energy Agency (IEA) and the Rocky Mountain Institute (RMI).OPGEE is designed to simulate upstream GHG emissions across four LCA stages. Exploration & Development, Production, Surface Processing, and Transportation. It can model various production methods such as waterflooding, gasflooding, steamflooding and gas lifting, along with common surface operations such as venting, flaring, and water & gas reinjection or disposal [8][9][11] [2].The latest iteration, OPGEE v3.0c, includes significant enhancements, particularly in the fugitive emission model. These improvements ensure a better representation of the GHG emissions from all oil and gas operations. A notable advancement in this version is the incorporation of a component-level fugitive model supported by a comprehensive database of component-level activity and emissions measurements [10].To our knowledge, this is the first time an LCA model of such an extent was built entirely as a PGM.

Methodology Overview

We propose to model the carbon intensity (CI) of an oilfield's operation from extraction to refinery. The boundaries of the system are represented in Figure [1]

Figure 1: Definition of oil and gas system's boundary for LCA methodolog

The model operates under the assumption that three fluids are present in the system:

1. Oil Stream: Oil is produced from the reservoir, passes through a downhole pump, and then is transferred to a separation process where oil, gas, and water are separated. Separated oil is stored in the crude oil storage tank and transported outside the system's boundary for further processing or shipping.
2. Gas Stream: Gas from the separation process is directed to flaring, where excess gas is burned off, and to venting, where gas is released directly into the atmosphere without being burned.
3. Water Stream: Water from the separation process is sent to a water treatment facility. After treatment, the water is either disposed of or used in water injection processes, likely for enhanced oil recovery or to maintain reservoir pressure.

We represent the LCA model as a PGM. PGMs([7]) are a class of statistical models that represent conditional dependencies between variables using a graphical structure that allows reasoning and inference under uncertainty.Key properties of PGMs include:

1. Graphical representation: PGMs employ graphs to visually depict the relationships between random variables. A node in the graph represents each variable, and the edges connect nodes with direct dependencies.
2. Probabilistic framework: PGMs utilize probability theory to quantify the strengths of these dependencies. Probability distributions are associated with each variable and edge, representing the likelihood of different values given the values of its parents or children.
3. Directed acyclic graphs (DAGs): restricts the graph structure to ensure that no directed cycles exist. This constraint ensures that there is no circular dependency among variables, making the model well defined and computable.

A PGM is a knowledge integration tool deployed to manage uncertainty and support decision-making processes. It can also be considered as an intuitive graphical representation of dependencies relationship between different indicators and variables. These dependencies are quantified using probability, hence making possible to use PGMs to compute the probability of an event given evidence on the other variables studied, which could be both hard (data) or soft (expert knowledge) evidences. TKRISK? (www.teokononda.com) is a Bayesian network framework that can be used to build risk models. The graph-based LCA model (OPGM) is thought of and built as a risk model. Figure [2] shows a graphical representation of the model.

Figure 2: Graphical representation of LCA-OPGM main modules and their relationships (TKRISK?)

The nodes presented in Figure [2] are actually collapsed groups. We present each of these groups over the next sections.

Benchmark: Global oilfields

In recent years, several efforts have been made to establish a global benchmark for oilfield GHG emissions. Notable among these efforts are three publicly available databases, which publish emission metrics such as Carbon Intensity (CI) and emissions for an exhaustive list of over 50,000 oilfields worldwide. These databases, which include The Archie Initiative, Fossil Fuel Registry, and OCI+, base their emissions computations on open-source lifecycle assessments (LCA) such as OPGEE. Their approach has been validated by multiple studies published in reputable journals ([4], [3], [5]).

Validation of the PGM Model

We have conducted a systematic comparison of GHG analytics generated by our Oilfield Pollutant Graphical Model (OPGM) with those produced by the most recent reference model, OPGEE v3.0c, on a benchmark of global oilfields. The results of this comparison are illustrated in Figure [3].

Figure 3: Cross plot comparison of CI (left), Field Emissions (middle), and upstream emissions (right) obtained from OPGEE v3.0c and the graph-based model (OPGM) for 1150 benchmark oilfield

The comparison indicates that all metrics align closely with those obtained from the reference model. To our knowledge, the PGM model can generate results that are as accurate as those of the latest version of OPGEE, v3.0c. As an example, the root mean square error (RMSE) on the average CI is of 6.95 kgCO2eq/BOE. This error lies within the measurement error made on oifields using state of the art sampling technologies ([10]). The amortized run-time for analyzing 1150 fields using the OPGM model is less than 1 second, compared to over 3 hours using OPGEE v3.0c.

Conclusion

This paper presents a significant advancement in Life Cycle Assessment (LCA) for oilfield operations through the development of the Oilfield Pollutant Graphical Model (OPGM). Our work demonstrates how Bayesian Networks can effectively model the complex interactions and uncertainties inherent in assessing the environmental impact of oil and gas production.The OPGM model, leveraging the capabilities of the TKRISK? Bayesian network framework, not only retains the essential functionalities of the latest OPGEE iteration (v3.0c) but also enhances them. By integrating this model into existing LCA practices, we have paved the way for more precise and faster environmental impact analyses. This is crucial for policymakers and industry stakeholders who seek to better understand and manage the environmental footprints of oilfield operations.Comparative analyses against a benchmark of over 1000 fields reveal that OPGM stands out as a scalable alternative to current LCA tools. It facilitates advanced evaluations, including sensitivity analysis and uncertainty quantification, thereby offering a more comprehensive understanding of the environmental impacts of oil and gas operations.In summary, the OPGM model represents a paradigm shift in LCA for oilfield operations, providing a more robust, intuitive, and efficient tool for environmental impact assessment. Its adoption could significantly contribute to the global efforts in managing and mitigating the environmental effects of the energy sector, aligning with broader sustainability goals.

References

[1]

Pearl, J.. Probabilistic reasoning in intelligent systems: Networks of beliefs. Morgan Kaufmann Publishers, Inc., 1988.

[2]

Masnadi, Mohammad S. and others. Global carbon intensity of crude oil production, pages 851--853. Science, 2018.

[3]

Masnadi, Mohammad S. and others. Carbon implications of marginal oils from market-derived demand shocks. Science, 2021.

[4]

Dixit, Yash and others. Carbon intensity of global crude oil trading and market policy implications. Nature Communications, 2023.

[5]

Liang, Jing and others. Carbon intensity of global crude oil refining and mitigation potential. Nature Climate, 2020.

[6]

International Energy Agency. Emissions from Oil and Gas Operations in Net Zero Transitions, 2023.

[7]

Koller, Daphne and Friedman, Nir. {Probabilistic Graphical Models: Principles and Techniques}. MIT Press, 2009.

[8]

Brandt, A.R.and Masnadi, M.S. and Rutherford, J.S. and El-Houjeiri, H.M. and Langfitt, Q. and Vafi, K. and Yulia, C. and Duffy, J.. {OPGEE v3.0b User Guide And Technical Documentation}, 2022.

[9]

El-Houjeiri, Hassan M. and Brandt, Adam R. and Duffy, James E.. Open-Source LCA Tool for Estimating Greenhouse Gas Emissions from Crude Oil Production Using Field Characteristics, pages 5998-6006. Environmental Science & Technology, 2013.

[10]

Jeffrey S. Rutherford and Evan D. Sherwin and Arvind P. Ravikumar and Garvin A. Heath and Jacob Englander and Daniel Cooley and David Lyon and Mark Omara and Quinn Langfitt and Adam R. Brandt. Closing the methane gap in US oil and natural gas production emissions inventories. Nature Communications, 2021.

[11]

Mohammad S. Masnadi and Patrick R. Perrier and Jingfan Wang and Jeff Rutherford and Adam R. Brandt. Statistical proxy modeling for life cycle assessment and energetic analysis, pages 116882. Energy, 2020.