Introduction

Systems-oriented analysis can help researchers gain a deeper understanding of sustainability issues (Sterman 2012; Levin et al. 2013; Liu et al. 2015; Preiser et al. 2018; Biggs et al. 2021). Analysts are increasingly recognizing that many environmental processes are fundamentally shaped by human activities and technologies mainly developed since the advent of the industrial revolution (Crutzen 2006; Steffen et al. 2007). Many of the types of physical relationships involving societal activities and biophysical processes that are relevant to sustainability are complex, dynamic, and changing over time (Sterman 2012). Therefore, a central activity of systems-oriented analysis for sustainability is to better examine and understand such relationships. Related processes are often characterized by nonlinearities, multiple feedback loops, and time delays in impacts, and many of these dynamics are currently poorly understood (Levin et al. 2013). A systems perspective can help analysts to build generalizable knowledge that can inform actions aimed to move societies toward greater sustainability. To support such sustainability transitions, researchers need to better understand sustainability through integrated research drawing on multiple disciplines and perspectives.

Researchers have developed a variety of systems-oriented frameworks related to sustainability analysis. These frameworks specify ways of conceptualizing systems and guide researchers toward identifying relevant elements, relationships, and processes (Biggs et al. 2021). Clark and Harley (2020) highlight sixteen interdisciplinary research programs that have shaped the field of sustainability science, including earth system governance, resilience thinking, and socio-technical transitions. Looking across these sixteen programs, they synthesize a high-level integrative framework of elements and relationships found to be important in sustainability science research. Binder et al. (2013) summarize ten prominent frameworks, assessing the ways in which interactions and dynamics are treated in each. Interactions relevant to sustainability have been characterized as taking place within social–ecological systems (Ostrom 2009; Levin et al. 2013; Colding and Barthel 2019), social–environmental systems (Eakin and Luers 2006), coupled human–natural systems (Liu et al. 2007; Chen 2015), human–environmental systems (Manuel-Navarrete et al. 2007), socio-technical systems (Markard et al. 2012), production–consumption systems (Lebel and Lorek 2008), and/or engineering systems (de Weck et al. 2011; Selin and Friedman 2012). These different system terms reflect variations in how researchers have conceptualized system components, and these components are understood to interact in multiple ways.

Despite the multitude of systems-oriented analytical frameworks, much empirical research on sustainability-relevant issues does not apply any of these frameworks (Selin 2021). This is the case both for research seeking to address practical challenges, and research which aims to build generalizable theories about how complex, adaptive systems operate. One reason for this may be that existing frameworks are not seen as accessible to those not already embedded in research communities that typically use them (Selin 2021); the different research communities that have developed and applied different systems-oriented frameworks are often fragmented into distinct traditions (Clark and Harley 2020). Existing systems-oriented frameworks also vary with the degree to which they set out a clear procedure for empirically grounded analysis using the framework (Binder et al. 2013). The need for developing additional research tools and methods has been recognized as a critical challenge to facilitate further empirical systems-oriented analysis and expand knowledge production relevant to advancing sustainability. For example, Elsawah et al. (2020), summarizing eight grand challenges for modeling socio-environmental systems, argued that further advances in multi-method approaches are needed to advance systems-oriented analysis.

In this article, we outline a new systems-oriented analytical framework—the human–technical–environmental (HTE) framework—and detail a matrix-based approach for its application. The article provides a synthesized guide to how researchers can apply the HTE framework that was first presented in Selin and Selin (2020). In particular, it builds on and extends the discussion in Selin and Selin (2020) on system component attributes, and further illustrates how insights from studies using the HTE framework can inform research and action on sustainability transitions as well as assist fruitful theory building. The HTE framework and the associated matrix-based approach are designed to provide researchers from different fields of study, including quantitative and qualitative traditions, a shared roadmap for empirically grounded interdisciplinary research on systems relevant to sustainability. Through such research, applications of the HTE framework using its matrix-based approach can generate empirical knowledge of much relevance to practical efforts focused on advancing sustainability transitions and contribute to the building of more advanced middle-range theories related to sustainability systems and sustainability transitions.

The rest of the article is divided into four sections. The next section describes the HTE framework and its matrix-based application through the process of identifying system components, studying interactions among these components and their outcomes, and examining results of past interventions while also considering opportunities for future interventions with the aim of advancing sustainability. The following section demonstrates how researchers can use the HTE framework and the matrix-based approach through an analysis of a system involving coal-fired power plants and mercury pollution. The subsequent section illustrates how researchers can use analysis based on the HTE framework to draw insights of relevance to sustainability science, highlighting three insights based on the analysis of the coal-fired power plants and mercury pollution system. The article concludes with a discussion of how the HTE framework can be further developed and applied to analyze a wide range of sustainability challenges in support of societal efforts to transition to greater sustainability and the development of middle-range theories relevant to sustainability science.

Applying the HTE framework through a matrix-based approach

The HTE framework is coupled with a matrix-based approach for empirically grounded analysis of systems relevant to sustainability. This combination is designed to provide researchers a way of conceptualizing systems relevant to sustainability that connects with a broad set of research methods, including both qualitative analysis from the social sciences and quantitative modeling typically used in natural science and engineering research. The HTE framework can be used to identify key components of a sustainability-relevant system. Its application through a matrix-based approach sets out a structured process for analyzing interactions among these system components and identifying potential interventions that could alter system components and/or their interactions. This structured process for using the HTE framework can also facilitate comparative analysis of different sustainability issues.

We designed the HTE framework to help researchers structure empirical data collection by providing five distinct categories into which individual components can be put for the purpose of systems-oriented analysis. The HTE framework draws its name from three categories of material components—human, technical, and environmental—that are critical to understanding many relationships between societal activities and biophysical processes that are of high relevance to sustainability science analysis. Institutions and knowledge, two categories of non-material system components, provide the context within which the material components interact. The application of the HTE framework and matrix-based approach progresses through four steps: (1) identifying system components and their attributes; (2) assessing interactions among system components; (3) specifying interveners and examining past interventions while also considering future potential interventions; and (4) drawing sustainability-relevant insights from the earlier analysis. This section summarizes the four steps of the HTE framework’s application.

Analyzing a problem from a systems perspective requires deciding which components to include and which to leave out of the system description. If everything is described as linking to everything else, identifying interactions of material system components can quickly devolve into an intractable analytical problem where the selection of too many components prevents meaningful empirically grounded analysis. The first step of applying the HTE framework is thus to identify the most important material human, technical, and environmental components together with the non-material institutional and knowledge components of a particular system for the purpose of sustainability-oriented analysis. The individual material components are the building blocks whose presence and interactions determine the state, structure, and function of a system, and each such component has time-dependent attributes that can change as a result of self-interactions or interactions with other material components. The non-material components also have their own specific attributes.

It is important to carefully select individual material and non-material components that most appropriately help analysts to understand important system dynamics related to sustainability. The total number of components needs to be large enough to capture these dynamics, but they also need to be few enough to allow for applied analysis of system operations and outcomes in practice. An analytically useful system description may include on the order of 20–30 components across all five categories (Selin and Selin 2020), but the number of individual components in each category as well as the total amount of system components is likely to vary on a case-by-case basis. As an analyst collects empirical information related to a specific case that they want to study from a systems perspective, thinking about which category-specific system components belong allows for the clustering of similar types of components into pre-defined categories.

Figure 1 summarizes the five categories of system components connected to the HTE framework and identifies some relevant generic attributes of each system component category. Human components are individual people or groups of people living in different places and under varying circumstances. They have physical attributes such as genetic conditions and age and social attributes such as education and wealth. Technical components consist of the built environment in the form of infrastructure or material artifacts of human society. These have attributes related to their mass, quantity, and operational efficiency. Environmental components consist of Earth’s life-support systems and the biosphere, including non-human living organisms in aquatic and terrestrial ecosystems. Attributes of environmental components include their physical properties and biological characteristics. Institutional components are social structures of rules, norms, and shared expectations, each with individual attributes related to their scope and stringency. Knowledge components involve information about other system components and their interactions and include attributes related to awareness and data availability. Similar to selecting only the most central individual system components, it is helpful to include only those attributes that are most relevant to the analysis.

Fig. 1
figure 1

Five sets of system components (sample attributes listed in parentheses)

Full size image

Some attributes of a system component include information about its spatial location and/or spatial extent that can be identified and measured relative to geographical distances or political boundaries on scales ranging from local to global. People included as human components may be community members who live within a small shared location such as a particular neighborhood or a municipality, but people in the form of consumers of particular products can be spread across multiple geographical regions and national jurisdictions. A technical component can be a single stationary point source such as a particular industrial plant, or it can be an extracted or manufactured material that may be transported across the world through both commercial trade and environmental processes. Environmental components may be the atmosphere, the oceans, or fish in a local lake. Institutions can come in the form of local practices and national laws as well as global treaties. Knowledge components such as human health information related to a particular toxic substance can be both highly localized and diffused all over the world.

System components can be defined at different levels of detail. For example, if it is analytically valuable to distinguish between different groups of workers, each group can be treated as a separate human component. If that distinction is not empirically relevant, all workers can be aggregated into a single component. In some instances, it may be useful to treat individual industrial plants as separate technical components, but all industrial plants can also be combined into one aggregated component if there is no analytical need to make such a separation. Sometimes, it may be helpful to treat different fish species as separate environmental components, but they can also be combined into one component if the greater level of detail is not analytically relevant. In some studies, it may be analytically useful to identify different pollution laws as separate institutional components, but for others it may be sufficient to treat all laws as one component. For some analyses, it may be necessary to identify data about the toxicity of different chemicals as separate knowledge components, but for other studies such data can be grouped into a single component.

The second step of applying the HTE framework uses a matrix as a heuristic tool to document and examine how the material system components that were identified in the first step interact in the context of institutions and knowledge. This is done by constructing an interaction matrix. Engineers and some systems analysts use matrices to better visualize, understand, and manage connections between system components (de Weck et al. 2011; Eppinger and Browning 2012). Matrices document the interactions of material system components, accounting for both one-way and two-way interactions. In other literature, conceptual diagrams in which different aspects of systems are connected with boxes and arrows are commonly used to visualize interactions (e.g. Ostrom 2009). A matrix presents the same information as box-and-arrow diagrams, but in a different structure. Compared to box-and-arrow diagrams, a matrix offers a more straightforward way to show the total number of individual components and study their interactions, including across different systems using a common conceptual structure. Relatedly, matrix approaches can be used to conduct network analysis of interdependencies, as called for by analysts of sustainability issues (Bodin et al. 2019).

While all five component categories have individual components with specific attributes, as discussed above, the construction of the interaction matrix involves taking into consideration only the attributes of the material components—that is, those that change over time as a result of interactions among human, technical, and environmental components against a background of existing institutions and knowledge. The attributes of the institutional and knowledge components remain constant for each implementation of the interaction matrix (but can be altered over time as a result of interventions). Figure 2 shows an illustrative interaction matrix for human, technical, and environmental components. Institutions and knowledge, which set parameters for interactions, are included in relevant matrix boxes. The total number of individual components can vary across system descriptions, and there may also be a different number of human, technical, environmental, institutional, and knowledge components within a single system. The matrix is read row first and column second. Figure 2, for illustrative purposes, includes one human component (H1), two technical components (T1, T2), three environmental components (E1, E2, E3), four institutional components (I1, I2, I3, I4), and five knowledge components (K1, K2, K3, K4, K5).

Fig. 2
figure 2

Sample interaction matrix

Full size image

Reading the matrix row first and column second provides directionality: a shaded box in a row and column in Fig. 2 indicates that the component in the row influences the component in the column. Thus, the shaded square in the first row and second column indicates that human component H1 (for example, factory workers) influences technical component T1 (like an industrial plant) by providing their labor. Shaded squares along the diagonal (e.g. in row T1 and column T1) indicate self-interactions of an individual component (the operation of the industrial plant). Institutional and knowledge components provide context for interactions of material components and are listed in the corresponding squares. For example, institutional component I3 (e.g., a local law or policy) governs the interaction where human component H1 influences environmental component E3 (which could be a local water reservoir); this may involve a local law or policy regulating human uses of water sources. Knowledge component K3 (which could be awareness about methods of pollution control) affects the interaction where technical component T1 influences T2 (which could be where an industrial plant influences another technical component T2, such as a pollution control device).

Interactions among material system components can change their relevant attributes, which were identified in the first step of selecting the system components. Interactions between two different material components (which can be of the same type or different types) change the attributes of the interacting components. For example, an interaction between a group of factory workers (H1) and a water reservoir (E3) as illustrated in Fig. 2 takes place when workers drinking water from the reservoir affect water availability (interaction of H1 with E3). This is affected by the existence of the local water law or policy (I3), which can influence decisions by factory workers on how much water they consume. The reverse interaction also occurs, whereby drinking reservoir water may also expose the factory workers to hazardous substances in the water (having a health effect, interaction of E3 with H1). The interaction can be characterized using knowledge of the benefits and risks (K5) of drinking water from the reservoir. Attributes can also change as a result of self-interactions that involve one component. For example, in the T1–T1 interaction in Fig. 2, the industrial plant’s attributes (which might include its output or its efficiency) can change as a result of its operation over time, in the context of a local law or policy (I3).

Figure 3 shows the qualitative descriptions of different types of interactions for a generic system; the bold text represents types of interactions that are present in the detailed matrix in Fig. 2. Institutions and knowledge are included as shaded background rectangles. Human components impact each other, where people and groups interact with each other (H–H). People interact with technical components by using technologies and products (H–T), and people interact with environmental components by harvesting resources and discharging pollutants (H–E). Technical components influence human components when technologies and products provide benefits or impose costs, including to their health (T–H). Some technical components interact with each other in autonomous configurations, such as when an emissions control technology interacts with other factory equipment (T–T). Technical components influence environmental components by leading to, or preventing, discharges of pollutants to ecosystems (T–E). Environmental components in the form of ecosystems provide services and resources, but degraded ecosystems can cause damage to people’s health and livelihoods (E–H). Environmental components can impact the performance of technologies, such as when the temperature of water affects the operation of cooling technologies in power generation (E–T). Ecosystem elements interact with each other such as when pollution is transported across different environmental media (E–E).

Fig. 3
figure 3

Generic qualitative interaction matrix for a human–technical–environmental system

Full size image

Once the interaction matrix has been constructed, researchers can use it to trace pathways of interactions across material components: pathways contain two or more linked interactions across human, technical, and/or environmental components (with these interactions taking place against the backdrop of existing institutions and knowledge). Such pathways of linked interactions can be used to trace causality, identifying which upstream material components affect other material components and their attributes downstream. Thus, when tracing causal influence forwards, the purpose is to look at how interactions between material components move forward chronologically through the matrix. The interaction matrix can also be used to trace pathways going backward. Such backward tracing is used to look at connections for causal attribution of effects, by identifying which previous sets of interactions caused a particular observed outcome. Figure 4 illustrates the kinds of different pathways that are found in Fig. 2, the sample interaction matrix for a human–technical–environmental system.

Fig. 4
figure 4

Illustration of interaction pathways related to the sample interaction matrix in Fig. 2

Full size image

The pathway diagram in Fig. 4 is different from a traditional box and flow diagram in which each box represents a component—the boxes in Fig. 4 illustrate the interactions with arrows connecting common components. This highlights the central focus on interactions in the HTE framework. For interactions that involve at least two components, the pathway diagram in Fig. 4 illustrates many of the types of pathways that are identifiable through the matrix-based approach. For example, two different interactions directly influence environmental component E3: those with human component H1 and environmental component E2. Thus, changes in the attributes of E3 can have multiple causal determinants. The interaction of human component H1 with technical component T1, followed by the self-interaction of T1, has multiple subsequent impacts, including on technical component T2 and environmental components E1 and E2, and then E3. Figure 4 also illustrates the reciprocal interactions and feedback loops. H1 and E3 affect each other, and H1 is also affected by the pathway of interactions involving T1, T2, E2, and subsequently E3.

It is often analytically helpful to begin with, or construct in parallel with a detailed matrix, a qualitative system description of material components and interactions such as the one in Fig. 3. If analysis using the HTE framework involves quantitative data or modeling, each interaction among material components can be considered as a separate function. Different types of quantitative models can be applied in the context of the HTE framework. These could include science and engineering models that capture stocks and flows of materials (such as global and regional dispersion of air pollutants), economic models that simulate changes in supply and demand of a particular resource or product in monetary terms, or agent-based models which capture interactions among different groups of humans in the context of varying resource availability. Further, the HTE framework provides an overarching analytical structure where different kinds of models can be used in parallel and be connected qualitatively, even when quantitative integration of these different models is not feasible.

The interaction matrix describes changing attributes of human, technical, and environmental components under a given set of rules shaped by the institutional and knowledge components at a specific time. The system’s evolution over time and geographical space—encoded in the value of each attribute—is determined by the interactions of the components. Temporal scales can be assessed by examining the length of time over which interactions occur, including interactions within an individual component, and can be measured according to specific rates. These rates can be expressed by quantities such as characteristic times or half-lives. Interactions can result in components changing their physical location, and physical scales of a problem can be analyzed by comparing the spatial attributes of the different components that interact with each other. The interactions, as evidenced by evolution of attributes of material system components as described by the interaction matrix, can exhibit complex, nonlinear behavior. Individual interactions can be of any functional form (e.g., linear, quadratic or exponential), and this form will depend on the rules governing the interactions (described by the institutional and knowledge components). The combination of these interactions occurring simultaneously can also illustrate system-wide nonlinearities and complexity.

The third step in applying the HTE framework involves evaluating how an intervener can change system behavior toward greater sustainability. This involves the identification of specific interveners. Interveners may include individuals, networks, governments, or organizations (including the private sector). Interventions can be considered either retrospectively (examining interventions that already have occurred) or prospectively (using the system as a model to examine potential for change). Interveners can target an individual human, technical, or environmental component by introducing a new material component or eliminating an existing one, or focus on changing interactions among two or more material components. Interventions may include introducing or changing institutions and/or knowledge components and their attributes that affect any of the material components or their interactions. Information related to interventions can be used to construct an intervention matrix. Similar to the way that interactions were described in the interaction matrix in Fig. 3, interventions affecting human, technical, and environmental components are aggregated for illustrative purposes in the intervention matrix in Fig. 5.

Fig. 5
figure 5

Generic qualitative intervention matrix for the human–technical–environmental system

Full size image

Interveners can target people and interactions among people (H–H). This can be done by changing how, for example, groups of workers engage with each other. Interveners can change human–technical interactions by targeting the way people use technology (H–T). This may involve introducing a new production method. Interveners can modify the impact of people on the environment (H–E). This can be done by new regulations, for example, those that prohibit discharges of toxic materials into waterways. Interveners can address how technological components impact people, such as by mandating worker protection measures (T–H). Interveners can alter the properties and operation of technologies by installing pollution prevention technology in an industrial plant (T–T). This can change the impact of technologies on the environment by affecting emissions and releases to ecosystems from point sources (T–E). Interveners can alter how ecosystems impact people (E–H). They can do this by, for example, diffusing knowledge about the risks of exposures to hazardous substances in the environment. Interveners can alter the limits that the environment poses to the operation of technologies (E–T). To do this, they could restrict the use of natural resources such as fossil fuels or minerals in certain technologies, or alternately develop new knowledge that enables technologies to operate under a broader range of environmental conditions. Interveners can also change both the structure and function of environmental components (E–E). They may do this by modifying ecosystems and related biological processes, for example, to remediate environmental damages in contaminated sites.

All of the interventions in Fig. 5 are described in qualitative terms, but the intervention matrix, similar to the interaction matrix, can also be used in quantitative systems modeling. Simulating interventions in a quantitative modeling framework generally requires introducing or subtracting components, or modifying the functional form of interactions between material components by changing or adding institutional or knowledge components. In a quantitative analytical approach, where the interaction matrix quantifies the rate at which the attributes of components change through time, the intervention matrix could describe increases or decreases in that rate of change. In mathematical terms, this corresponds to a type of second derivative. In systems analysis, mathematical calculations of this type are often used to determine system behaviors, timescales, and properties. A quantitative intervention matrix might for example be used to describe how quickly and by how much the rate of emissions from an industrial point source is reduced through the application of pollution prevention technologies or other forms of operational changes, as well as how these reductions impact environmental deposition and human health in locations both near and far from the point source.

A particular person or group of people can be treated as a human component and/or as a potential intervener. For example, a factory owner can be included as a human component that interacts with other material components under a given set of institutions and knowledge. At the same time, a factory owner can be treated as an intervener, if the analysis wishes to capture their ability to intentionally and directly intervene by altering other material and non-material components and/or their interactions. The factory owner can intervene by changing working hours and conditions in the factory in ways that affect the health of factory workers, or installing new technology that results in fewer emissions of hazardous substances to the environment. For analysis related to interventions, it is often analytically helpful to draw lessons from past interventions and to consider new options for future interventions; past interventions in a system may be included in a present-day interaction matrix, or treated as an intervention in an analysis designed to evaluate the effectiveness of existing strategies. Such analysis of interactions and past interventions using interaction and intervention matrices and pathway diagrams in turn can help analysts to think constructively and critically about practical opportunities for future interventions.

In the fourth and final step of the application of the HTE framework, analysts can draw sustainability-relevant insights from the earlier analysis of system components, interactions, and interventions. We differentiate between insights that are case specific, focused on improving understandings of system operations and how specific system behavior can be practically altered in ways that inform action toward sustainable development, and those that can aid in the building of more generalizable knowledge and theory. Selin and Selin (2020) group individual insights into the three thematic areas of systems analysis for sustainability, sustainability definitions and transitions, and sustainability governance. Other thematic areas are also possible, as the purpose is to synthesize empirically grounded insights that can help advance further analysis and/or support societal efforts to advance sustainability.

Applying the HTE framework: coal-fired power plants and mercury pollution

In this section, we apply the HTE framework to a specific case, analyzing a system involving coal-fired power plants and mercury pollution. Qualitative research has examined the negotiations and early implementation of the 2013 Minamata Convention on Mercury, which controls mercury emissions from coal-fired power plants (Selin 2014; Selin et al. 2018; Uji 2019). Quantitative research has linked models of human activity to simulations of the environmental fate and transport of mercury emissions from coal burning. Giang and Selin (2016) used projections of future mercury emissions under US domestic law and Minamata Convention regulations of human exposure in the USA, and evaluated the health impacts on people living in different US regions using an economic model. Chen et al. (2019) examined the impacts of economic activities and trans-provincial trade in goods and services on the health impacts of related mercury emissions in China. Quantitative research has also explored linkages been mercury emissions from coal-fired power plants and climate change. Mulvaney et al. (2020) used a computable general equilibrium model of the Chinese economy to explore how implementation of the 2015 Paris Agreement on climate change and the Minamata Convention affect mercury emission reductions and deposition. Compared with these previous studies, using the HTE framework to analyze mercury emissions from coal-fired power plants allows for a more integrated analysis with a focus on advancing sustainability.

The application of the HTE framework and the matrix-based approach offers researchers an opportunity to structure an empirically grounded systems-oriented study of a sustainability issue. Below, we detail the stepwise process of identifying system components and examining interactions and interventions related to mercury emissions from coal-fired power plants. Mercury is a major pollutant that is dispersed in the environment globally (United Nations Environment Programme 2019) and linked to several of the United Nations Sustainable Development Goals, including good health and well-being (goal 3), affordable and clean energy (goal 7), responsible consumption and production (goal 12), and life below water (goal 14) (United Nations Development Program and Global Environment Facility nd). System components are highlighted in italics in the description below and summarized in Fig. 6, along with analytically relevant attributes for each material component. To differentiate these components from the sample system components above, each system component is labeled beginning with Hg- (for mercury); that is, the first human component would be Hg-H1, etc. Analysis related to interactions and interventions is also discussed below and summarized in Figs. 7, 8, and 9. This combined analysis of components, interactions, and interventions draws from and extends on the more general discussion of point sources of mercury emissions in chapter five of Selin and Selin (2020).

Fig. 6
figure 6

System components for the coal-fired power plant and mercury pollution system with attributes of the material components in parentheses

Full size image
Fig. 7
figure 7

Detailed interaction matrix for the coal-fired power plant and mercury pollution system

Full size image
Fig. 8
figure 8

Interaction pathways in the coal-fired power plant and mercury pollution system. Green boxes highlight focal interactions discussed in text, and red boxes and text highlight selected interventions

Full size image
Fig. 9
figure 9

Intervention matrix for the coal-fired power plant and mercury pollution system

Full size image

Human activities have discharged mercury into the environment for millennia, but emissions to air increased sharply at the beginning of the industrial revolution. Mercury is emitted by a variety of anthropogenic sources to the atmosphere (environmental component Hg-E1 with mercury levels as an attribute). Coal contains trace amounts of mercury, and global mercury assessments that quantify emissions to air identify coal-fired power plants (technical component Hg-T1 with attributes of energy production, emissions, and lifetime) as a major source of mercury emissions (United Nations Environment Programme 2002; United Nations Environment Programme 2019). Historically, much mercury was emitted by coal-fired power plants in North America and Europe, but today Asia (and in particular China and India) are leading sources of mercury emissions (United Nations Environment Programme 2019). The continuous burning of coal takes place alongside the use of other energy sources (technical component Hg-T2 with attribute of energy production), which may include both fossil and non-fossil fuel sources. The use of pollution control technology (technical component Hg-T3 with attribute specifying its efficiency) can capture mercury in coal before it is emitted into the atmosphere. Such captured mercury (technical component Hg-T4 with attribute of its quantity) in fly ash must be subject to environmentally safe storage and waste disposal to avoid being discharged into the environment at a later date.

Decisions influencing energy generation from coal-fired power plants and other energy sources are made by producers and consumers of energy (human component Hg-H1 with attributes specifying levels of supply, demand, and access). Many of these decisions are shaped by supply and demand dynamics of energy markets (institutional component Hg-I1). In turn, national and local laws and regulations (institutional component Hg-I2) can control mercury emissions from coal-fired power plants. The application of pollution control technology depends on awareness of Best Available Technology options (knowledge component Hg-K1) for reducing emissions. The Minamata Convention on Mercury (institutional component Hg-I3) globalizes this approach to controlling mercury emissions by, for example, mandating the application of pollution controls to stationary sources such as coal-fired power plants. The Minamata Convention does not restrict coal burning per se, but its focus on reducing mercury emissions from coal-fired power plants creates a shared focus on fossil fuel burning with the Paris Agreement on climate change (institutional component Hg-I4).

Scientific knowledge of atmospheric transport and fate of mercury emissions and mercury pollution exposure pathways (knowledge components Hg-K2 and Hg-K3) has evolved significantly since the 1990s. Today, scientists recognize that different forms of mercury travel in the atmosphere both regionally and globally. As a result, some of the mercury emitted from coal-fired power plants deposits in ecosystems near coal-fired power plants, while the rest may reach ecosystems far from coal-fired power plants (environmental components Hg-E2 and Hg-E3 with attributes specifying mercury levels). As a result, mercury poses health risks to both people living near coal-fired power plants and people living far from coal-fired power plants (human components Hg-H2 and Hg-H3 with attributes characterizing their health). The form of mercury that often poses most risk to humans and animals is methylmercury, a dangerous neurotoxin. Methylmercury is present in mainly aquatic ecosystems and increases in concentrations up aquatic food webs, where humans and other large predators who are at the top of such food webs are most at risk of suffering adverse health effects from its exposure.

Based on the above selection of individual components in each of the five categories of human, technical, environmental, institutional, and knowledge, we construct an interaction matrix to identify and examine interactions among the material components important to sustainability in the context of the identified institutional and knowledge components. Figure 7 presents this detailed interaction matrix, which follows the generic model of Fig. 2. Because the interaction matrix is read row first and column second, the shaded boxes in Fig. 7 indicate that an interaction occurs where the material component in the row influences the material component in the column. Interactions modify the attribute or attributes of the target component listed in the black text in the box. Red text in brackets in each box identifies the institutional and knowledge components that affect the functional form of this interaction. For example, producers and consumers of energy (Hg-H1) influence energy production in coal-fired power plants (Hg-T1) as a result of decisions impacting the supply and demand for energy, and this interaction is shaped by dynamics in energy markets (Hg-I1).

Drawing on the interaction matrix in Fig. 7, Fig. 8 presents a flowchart of linked interactions. The causal chain in Fig. 8 represents a series of linked interactions starting with producers and consumers of energy interacting through energy markets (self-interaction of Hg-H1, affecting attributes of supply, demand, and access; box highlighted in green in Fig. 8). The ultimate impact is mercury in ecosystems affecting people, primarily through eating fish containing methylmercury. This is reflected by the interactions between Hg-E2 and Hg-H2, and Hg-E3 and Hg-H3, respectively, for people living near and far from coal-fired power plants, affecting attributes related to their health, also both highlighted in green in Fig. 8. Producers and consumers of energy influence both coal-fired power plants and other energy sources through energy markets (interaction between Hg-H1 and Hg-T1 and Hg-T2, affecting the attribute energy production). These different sources produce energy which benefits producers and consumers (Hg-T1 and Hg-T2 interact with Hg-H1, influencing their access to energy). Interactions whereby coal-fired power plants emit mercury to the atmosphere (interaction between Hg-T1 and Hg-E1, affecting the atmosphere’s mercury levels) where this mercury is transported to both nearby and far away ecosystems (interaction between Hg-E1 and Hg-E2, and Hg-E1 to Hg-E3, affecting mercury levels in ecosystems) are central in this series of linked interactions.

As shown in Fig. 8, not all mercury in coal is emitted to the atmosphere when coal is burned to generate energy. The operation of pollution control technology (Hg-T3 interacting with itself, affecting attribute of efficiency) captures some of the mercury in coal, preventing emissions to the atmosphere (Hg-T3 interacts with Hg-T4, influencing the attribute amount of mercury that is emitted to air; Hg-T3 interacts with Hg-T1, affecting the attribute mercury emissions from coal-fired power plants). However, if the captured mercury is later released to ecosystems near coal-fired power plants (interaction of Hg-T3 with Hg-T2) it starts cycling in the environment (Hg-E2 interacts with Hg-E1) where it is transported shorter or longer distances via the atmosphere (Hg-E1 interacts with Hg-E2 and Hg-E3). For example, re-use of fly ash captured from end-of-pipe control technologies in cement production can lead to captured mercury from coal-fired power plants being released to ecosystems and also ending up in the atmosphere, thereby affecting the health of people living both nearby and far from coal-fired power plants.

Analysis building on the interaction matrix can help identify interveners and interventions that can address mercury emissions from coal-fired power plants. This forms the basis for the intervention matrix shown in Fig. 9. National and local governments often play important roles as interveners by adjusting attributes of existing institutions or introducing new institutions related to mercury emissions and exposure as well as climate change mitigation. For national governments of countries that are parties to the Minamata Convention, the Conference of Parties to this treaty can take steps as interveners toward the strengthening of domestic laws and regulations by updating treaty-based rules and guidance on controlling mercury emissions from stationary sources. Private sector actors such as owners of industries that operate power plants can install mercury-capturing control devices to reduce damages to the environment and human health. Interventions can also be dependent on the generation, diffusion, and application of new knowledge, which highlights the importance of technical experts. Consumers may have the ability to purchase energy from different non-coal sources, and civil society groups as well as governments can run public information campaigns related to the reduction of methylmercury exposure from fish consumption.

The interventions shown in Fig. 9 affect either components or interactions, and these map onto the pathway diagram in Fig. 8. Where interventions alter interactions, these are illustrated by the boxes highlighted in red. Consumer changes in energy demand, in box H–T in Fig. 9, affect the boxes in Fig. 8 in which producers and consumers of energy influence production of energy by coal-fired power plants and other energy sources (interactions of component Hg-H1 with components Hg-T1 and Hg-T2, respectively). When owners of a coal-fired power plant improve pollution control technology this can affect its efficiency, a technical–technical interaction (Hg-T3 interaction with itself). Other interventions target institutional and knowledge components, which can affect multiple interactions, are also highlighted in red in Fig. 8. Experts can develop new knowledge on technology-based control options, such as best available technology options (Hg-K1). Non-governmental organizations can run information campaigns to reduce mercury exposure from eating contaminated fish, which affects knowledge of exposure pathways (Hg-K3). In turn, this may influence the interaction between mercury in ecosystems and humans (e.g. the interaction of Hg-E2 with Hg-H2, and interaction of Hg-E3 with Hg-H3). National and local governments can mandate the use of pollution control technologies and/or regulate levels of mercury emissions and releases through setting national and local laws and regulations (Hg-I2). The Minamata Convention Conference of Parties can influence discharges by changing treaty provisions on emissions and releases in the Minamata Convention (Hg-I3).

Drawing lessons from past interventions can help analysts consider options for future interventions. Evaluating impacts of past interventions requires accounting for previous changes in dynamics such as those that can be seen along the interaction pathway shown in Fig. 8. In the empirical case of mercury emissions from coal-fired power plants, some national and local governments have already introduced laws and regulations mandating the application of end-of-pipe control technologies targeting mercury emissions, and/or taken steps toward phasing out coal-burning while supporting other energy sources, particularly non-fossil energy sources, which reduce both carbon dioxide and mercury emissions. Both of these kinds of interventions are examples of past interventions that have reduced human health risks by reducing the amount of mercury that is emitted into air and converted to methylmercury through environmental processes. We choose to treat them as interventions in this illustration, as one purpose of our analysis is to help identify the ways in which such interventions might protect human health by changing interaction pathways such as those illustrated in Fig. 8.

Insights from the HTE framework analysis of the coal-fired power plant and mercury pollution system

The fourth step of applying the HTE framework involves drawing sustainability-relevant insights from the earlier empirically grounded analysis focused on system components, interactions, and interventions. Based on our application of the HTE framework to the coal-fired power plants and mercury pollution system, we identify three case-specific insights that are, taken together, more comprehensive than those found by previous individual analyses of mercury pollution from coal-fired power plants using other approaches and methods. Each of these case-specific insights suggests a broader insight that is of generalizable interest to the sustainability science community. First, the use of the HTE framework and the interaction and intervention matrices make clearer the different impacts of targeting “upstream” and “downstream” leverage points via particular interventions in a comparable way. Second, the matrix-based analysis shows how socioeconomic dynamics of energy markets directly connect to the ultimate effects of pollution control technologies, with influences on long-term pollution levels. Third, analysis using the HTE framework provides a way to formalize analyses of polycentric governance structures and how different institutions fit relevant biophysical processes and socioeconomic and political aspects. We discuss each of these insights below.

First, the HTE framework allows analysts to explicitly identify where a possible leverage point for changing system operations is located along a pathway of causal influence, and thereby better understand impacts, including potential feedbacks, of specific interventions. For mercury specifically, some “upstream” interventions identified in Fig. 8 address mercury discharges early in a causal pathway, while other more “downstream” interventions on mercury exposure focus on interactions located later in a pathway. Upstream interventions such as those targeting mercury emissions to air from coal-fired power plants through the introduction of pollution prevention technology may have different capacities to influence outcomes than do those that are taken further downstream such as issuing fish consumption advice to people eating seafood containing methylmercury. To protect the environment and human health from mercury, it is necessary to address mercury use, discharges, and exposure in a comprehensive manner (Selin 2011).

More generally, analysts (and decision-makers) increasingly acknowledge that upstream and downstream interventions often need to occur simultaneously to maximize their impact. This idea is related to the concepts of climate change mitigation and adaptation, with mitigation looking to reduce greenhouse gas emissions and adaptation referring to societal efforts to adjust to a changing climate. The HTE framework can be used to show how upstream and downstream interventions targeting different leverage points are connected in a single system. Switching “upstream” energy sources (interventions targeting interactions of component Hg-H1 with components Hg-T1 and Hg-T2) will eventually have a positive influence on human health by reducing exposure to methylmercury from food consumption at the end of the causal pathway. However, as each interaction has an associated timescale, this is not an action with an immediate effect, and may not lead to reduced human methylmercury exposure until at least several decades into the future based on timescales for energy transitions and environmental cycling of mercury. Correspondingly, interventions such as public awareness campaigns that provide dietary advice related to fish consumption (targeting Hg-K3) can have relatively immediate effects on human health, but do not affect pollution at the source. The HTE framework allows analysts to clearly visualize and potentially calculate simultaneous upstream and downstream interventions can have greater influence on human well-being in the near and longer terms than either approach in isolation.

Second, the HTE framework highlights the nuanced nature of the connection between end-of-pipe technology application and mercury emissions, by showing that this interaction is mediated by the dynamics of energy markets. Previous research considered the installation of end-of-pipe technology application on coal-fired power plants as an intervention that affected the amount of mercury emitted to the atmosphere and/or released to land and water (Rafaj et al. 2013; Giang et al. 2015). Separately, other analyses considered the effect of governments mandating the installation of mercury pollution capture technologies on plant closures—most of this literature relies on economics-based methods (Linn and McCormack 2019; Coglianese et al. 2020). However, analysis of the coal-fired power plant and mercury pollution system shows that these two interactions are connected such that energy markets mediate the impact of mercury emissions changes. Figure 8 illustrates this, as pollution control technology operation influences coal-fired power plants (interaction of Hg-T3 with Hg-T1), which also interact with other energy sources (interactions between Hg-T1 and Hg-T2), and the coal-fired power plants influence the atmosphere through emitting mercury (interaction between Hg-T1 and Hg-E1).

Analysis using the HTE framework can help identify trade-offs between the application of end-of-pipe pollution control technologies and longer-term dynamics of energy markets, illustrating how technical systems operate in the context of socioeconomic drivers. Tracing the pathway of linked interactions through Fig. 8 identifies the way in which this potential trade-off ultimately influences the harms of mercury pollution to human health. The installation of pollution control technology reduces the levels of mercury emissions, but installing such technology may also influence the lifetime of a coal-fired power plant (and thus the degree to which reduced mercury emissions continue into the future). Governments mandating the application of new pollution control technology may both reduce and extend the lifetime of coal-fired power plants. In the USA and Canada, implementation of advanced pollution control standards for coal-fired power plants in the 2000s and 2010s was reported to have resulted in earlier closure of some older plants (Sloss 2012; Storrow 2017). The opposite dynamic may exist in other regions and countries, including China and India—it is possible that the incorporation of advanced pollution control, for example on newly built coal-fired power plants, could lengthen their lifetime, extending the impact of subsequent mercury and carbon dioxide emissions to future generations. This kind of analysis is also relevant to studies of other pollutants from both stationary industrial sources and mobile sources running on fossil fuels.

These first two insights furthermore suggest how quantitative modeling using the HTE framework can support additional analysis. The first insight highlights the need to account for both upstream and downstream drivers of sustainability-relevant outcomes in modeling that supports causal inference. Attributing changes in mercury exposure should account for both changes in fish consumption and related advice as well as upstream efforts to reduce atmospheric emissions from major sources. The second insight draws attention to the need for modelers to incorporate the dynamics of both energy systems and energy markets into analyses, including rules and regulations that influence choices between coal, other fossil fuels, and non-fossil energy sources. A life cycle approach that tracks captured mercury, which is not subject to environmentally safe storage and disposal and could enter the environment and subsequently affect people all over the world, is also needed. Economic analysis of the dynamics of the energy system links to life cycle analysis as well as atmospheric fate and transport modeling of mercury—three different modeling methodologies that are currently often used separately and by investigators from different fields in analyzing mercury pollution.

Third, the HTE framework helps formalize analysis of polycentric governance and institutional design. Issues of indirect institutional linkages can be examined using the HTE framework, which may inform decision-making on mercury and related issues. Hg-I3 (the Minamata Convention) and Hg-I4 (the Paris Agreement) do not directly influence the same interactions. During the Minamata Convention negotiations, countries agreed that the treaty would not restrict coal burning per se (Selin 2014). Any such effort would have made it more difficult (if not impossible) to reach an agreement on a new global mercury treaty. This would have delayed the application of (stricter) mercury pollution control technologies in at least some countries. Thus, only targeting mercury emissions resulted in quicker negotiations and also allowed for major countries (including China, India, the USA, Japan, Indonesia, and South Africa) alongside the European Union to become parties to the Minamata Convention. However, an indirect connection between the Minamata Convention and the Paris Agreement results from the material interactions involving the installation of mercury-capturing technologies on coal-fired power plants (interactions between Hg-T3 and Hg-T1) that can influence the lifetime of a coal-fired power plant, with implications for both mercury and carbon dioxide emissions.

More generally, analysts can study polycentric governance structures by focusing on the numbers and types of institutional components at different governance levels that affect each material component and its interactions. In the coal-fired power plant and mercury pollution case, the Minamata Convention (Hg-I3) and national and local laws and regulations (Hg-I2) can affect four different interactions (Hg-T3 with Hg-T3, Hg-T3 with Hg-T1, Hg-T1 with Hg-E1, and Hg-T4 with Hg-E2). These involve, respectively, the operation of pollution control technology (by mandating applications); the capture of mercury by that pollution control technology (through efficiency standards); emissions to the atmosphere (by setting emission limits); and the potential release to ecosystems (by regulating the handling of fly ash). The content and stringency of national and local laws and regulations may vary across countries, and there can be different national and local institutions related to fishing and fish consumption. Analysis can also focus on the degree to which these institutions, individually and collectively, match (or ‘fit’) both biophysical dynamics of how mercury cycles through the environment and socioeconomic and political aspects of the mercury issue for the purpose of reducing the negative impacts of mercury exposure on human health. Such institutional analysis is also valuable in many other areas of environmental governance, including climate change and hazardous chemicals.

Finally, while relevant insights such as those above could be achieved by different analyses of the coal-fired power plant and mercury pollution system in isolation, an interdisciplinary benefit of the HTE framework is that analysis using a common framework can lead to these multiple insights simultaneously. This also helps to combine different areas of studies to strengthen sustainability analysis. The first insight links engineering-based and policy analysis through the identification of leverage points. The second insight combines socioeconomic analysis with technological forecasting. The third insight links governance-based studies of institutional design and effectiveness with natural science-based and public health analyses of environmental flows of materials and human health risks from exposure to hazardous materials. Importantly, analysis of mercury emissions from coal-fired power plants presented here using the HTE framework could be conducted by researchers from any of these disciplines (where there are clear benefits of bringing together researchers from multiple fields combining different areas of expertise) and further linked to quantitative and qualitative research methods within specific disciplinary traditions and areas of research.

Further application and development of the HTE framework

In this article, we outline a new systems-oriented analytical framework, the HTE framework, which can be used to empirically examine sustainability-relevant issues. The framework is specifically designed to be accessible both to researchers in the social sciences and to those who focus on sustainability issues from a natural science and engineering perspective. We further illustrate specific insights that can be gained through its application using a matrix-based approach, by applying it to a case study of coal-fired power plants and mercury pollution. In general, a benefit of applying the HTE framework is that it helps researchers better understand interactions between people, technologies, and the environment, how these interactions can both advance and hinder sustainability transitions, and opportunities for interventions that support societal changes toward greater sustainability. Many of these interactions between human activities and biophysical processes, as well as the design of interventions, involve relationships and processes that are highly complex, creating an urgent need to better understand such relationships and processes for the purpose of advancing sustainability. Here, we briefly discuss steps forward for further application of the HTE framework and how such research can contribute to theory development.

The HTE framework can be further applied and tested across a range of empirical cases to explore its utility for analyzing sustainability-related issues. To date, the most comprehensive application of the HTE framework has focused on mercury as both a commercial substance and a pollutant posing major environmental and human health problems, which the analysis in this paper builds upon (Selin and Selin 2020). However, the HTE framework has also started to be applied to other cases, both for research and teaching applications. These include single-use plastics (Christoff-Tempesta and Selin 2021); air pollution and agricultural residue burning in India (Maji et al. 2020); and water and infrastructure in the Indus Basin (Siddiqi et al. 2020). In particular, the HTE framework’s focus on material interactions in the context of institutions and knowledge suggests that it might be particularly suited to analyzing different types of material flows with strong technological components, including those of other pollutants, goods, nutrients, water, or energy. In addition, its application to examine place-based systems might result in insights distinct from those gleaned from studies using more traditional social–ecological systems framework, such as those used for studying common-pool resources (Ostrom 2009). All such further applications of the HTE framework could help more explicitly illustrate its advantages and limitations.

The HTE framework contributes to systems analysis for sustainability through its combination of three separate material component categories and two non-material component categories coupled with a matrix-based approach for identifying system components and examining interactions and interventions. The framework and the matrix-based approach addresses a gap between the large amount of empirical work conducted in particular on environmental challenges, and the lack of a systems approach to address important interactions, especially those involving institutions and knowledge components (Selin 2021). Through its focus on both material and non-material system components, the HTE framework can inform research on sustainability transitions. The HTE framework’s characterization of technology as a separate category of material system components resonates with much historical transitions-focused research. Specifically, the interaction and intervention matrices can be used to explain dynamics of change mechanisms, and to assess the impact of normative interventions including those that involve new institutions or knowledge that change the structure and function of a system.

Some authors in the transitions literature draw a distinction between transitions and transformations related to degrees of changes to structures and functions of sustainability-relevant systems, with the latter defined as more fundamental or radical change (Feola 2015). The HTE framework can be used in future work to evaluate and compare degrees of systemic change that result from different interventions, as interventions may simultaneously produce both smaller and larger changes to system operations (e.g. Maji and Selin 2022). The HTE framework can further be used to highlight whether changes are the result of intentional interventions by specific interveners, which allows for conceptual separation for evaluating actors’ normative goals. The HTE framework’s focus on leverage points provides analytical structure to efforts to compare transitions and transformations across different contexts (see e.g. Linnér and Wibeck 2021). In addition, Loorbach et al. (2017) highlight largely separate research strands in the transitions literature focusing on socio-technical, socio-institutional, and socio-ecological perspectives; the comprehensive nature of the HTE framework could help bridge analyses across these and other different research communities.

Additional application of the HTE framework can be facilitated by the fact that the matrix-based approach serves as a mechanism for linking qualitative and quantitative research and data. This has been identified as a priority in the transitions literature (Turnheim et al. 2015) as well as for advancing socio-environmental systems modeling (Elsawah et al. 2020). The matrix-based approach can assist such analysis. Incorporating institutional constraints in sustainability-focused modeling could provide insights to guide real-world decision-making, for example in identifying which scenarios applied in integrated assessment modeling are more feasible than others (Brutschin et al. 2021). Once an analyst has constructed the interaction and intervention matrices for a particular system, it is possible to use a variety of approaches to assess or simulate how each material component’s attributes change over time as a result of the interactions and interventions described in the matrices. Importantly, not all of the interactions or interventions need to be quantified for a quantitative model to be useful to assess how system attributes change over time.

The HTE framework can help a systems modeler account in a rigorous way for the components and interactions they are not able to quantify, without having to omit them entirely from the analysis. Analysts can also make choices depending on their objectives about what processes to endogenize in modeling as part of the interaction system, and what to evaluate as an intervention. For example, when applying the HTE framework to a hazardous substance such as mercury, an analyst may have an existing model of cycling among environmental and human components and be interested in how a certain technology could change concentrations. In this case, the application of a technology could be incorporated as an intervention, as interventions can provide a structure for developing model scenarios. In contrast, if the analyst were interested in simulating how new technologies emerge, for example to inform knowledge about innovation and transition processes, the analyst could construct an interaction matrix that included processes for developing innovations. Potential interventions could in that case involve interveners who develop new knowledge or set up new institutions for encouraging actions that might lead to innovations.

The use of the HTE framework also assists researchers in thinking systematically about issues of vulnerability, power, and equity with respect to interactions and interventions. The inclusion of human components puts a focus on varying levels of vulnerability and risks for different people and groups of people. With respect to mercury exposure as well as other cases of pollution and environmental degradation, impacts on human health and well-being are typically not uniform across all affected populations. Analysis involving selection of system components and detailing interactions can identify and highlight the most exposed and vulnerable peoples. Subsequent analysis using an intervention matrix can be used to identify potential interveners and the most influential leverage points—that is, those who have the power to change particular system components and how material components interact. Individual interveners may have separate capabilities. Different interventions are likely to have varying impacts on people’s health and well-being, including those who are most exposed and vulnerable to pollution and environmental degradation. Analysis focused on interventions can thus bring to the fore aspects of power and equity when considering the relative impacts and benefits to different peoples of specific interventions.

Studies using the HTE framework can inform development of middle-range theories related to sustainability systems and sustainability transitions. This involves using insights from case-based studies to formulate broader explanations relevant to a larger set of related cases (Merton 1968). In relation, Meyfroidt et al. (2018, p. 53) defines middle-range theories as “contextual generalizations that describe chains of causal mechanisms explaining a well-bounded range of phenomena, as well as the conditions that trigger, enable, or prevent these causal chains.” Systems-oriented frameworks such as the HTE framework can provide guidance for researchers hypothesizing about such causal mechanisms (Vos et al. 2021). This involves an attempt to “maximize the generalizability of insights across contexts, but without compromising the validity of such generalizations” by aiming for universal explanations (Bodin et al. 2019, p. 555). While some empirical research addresses particular relationships between human activities and biophysical processes, a dearth of analytical frameworks that allow for careful and systematic comparisons of these relationships across different empirical cases limits the utility of this research for further theory development (Selin 2021). In this respect, the HTE framework provides a tool for advancing middle-range theory building.