Abstract
Digitally enabled technologies are increasingly cyber-physical systems (CPSs). They are networked in nature and made up of geographically dispersed components that manage and control data received from humans, equipment, and the environment. Researchers evaluating such technologies are thus challenged to include CPS subsystems and dynamics that might not be obvious components of a product system. Although analysts might assume CPS have negligible or purely beneficial impact on environmental outcomes, such assumptions require justification. As the physical environmental impacts of digital processes (e.g. cryptocurrency mining) gain attention, the need for explicit attention to CPS in environmental assessment becomes more salient. This review investigates how the peer-reviewed environmental assessment literature treats environmental implications of CPS, with a focus on journal articles published in English between 2010 and 2020. We identify nine CPS subsystems and dynamics addressed in this literature: energy system, digital equipment, non-digital equipment, automation and management, network infrastructure, direct costs, social and health effects, feedbacks, and cybersecurity. Based on these categories, we develop a 'cyber-consciousness score' reflecting the extent to which the 115 studies that met our evaluation criteria address CPS, then summarize analytical methods and modeling techniques drawn from reviewed literature to facilitate routine inclusion of CPS in environmental assessment. We find that, given challenges in establishing system boundaries, limited standardization of how to evaluate CPS dynamics, and failure to recognize the role of CPS in a product system under evaluation, the extant environmental assessment literature in peer-reviewed journals largely ignores CPS subsystems and dynamics when evaluating digital or digitally-enabled technologies.
Export citation and abstract BibTeX RIS
1. Introduction
Many emerging environmental solutions and technologies rest on the ability to quickly gather, update, and process information via digital networks, unlocking efficiencies and providing service flexibility. Technologies that provide this information processing capability are generally termed information and communication technologies (ICTs). ICT capabilities are central to technology realms including smart grid, autonomous vehicle (AV) and building controls, and blockchain for transaction verification (Asad and Rehman Chaudhry 2017, Lee et al 2019, Sharma et al 2020). As ICT applications spread to infrastructure systems and societies throughout the world, their environmental impact has also become a matter of public interest, including for data center energy and material requirements, electric vehicle battery life cycles, and cryptocurrency energy demand (Koomey 2008, Dandres et al 2017, Arbabzadeh et al 2019, Masanet et al 2020, Cellan-Jones 2021).
In 2020, the response to COVID-19 brought with it a global shift from physical to cyber-connectedness as classrooms, gatherings, and many workplaces moved to virtual environments. As such, many 'digital winners' have emerged from the need to connect to both goods and services online (HBS 2021). In the literature on ICT impacts, a commonly cited figure suggests that the ICT sector contributes about 2% of global greenhouse gas emissions, similar to that of the aviation industry (Whitehead et al 2014, Bull 2015). Others have disputed this claim, or simply noted that drawing a rigorous comparison between the two sectors is fraught due to system boundary inequivalence (Belkhir and Elmeligi 2018). Either way, projections for both US and global energy systems highlight slow recovery ahead for aviation, while digital services are a central part of overall economic recovery (EIA (Energy Information Administration) 2021). The environmental implications of widespread ICT integration and use is thus an urgent concern (Masanet et al 2020).
Research and writing on this topic utilizes many different terms to describe the widening array of ICT applications, including connectedness, digitalization, digitally-enabled, autonomous, and smart. In this study, we use the term cyber-physical system (CPS) as an umbrella concept that captures all of these terms and is focused on the intersection of infrastructure, ICT, and human interaction (Rajkumar et al 2010). Our aim is to evaluate the state of environmental assessment methodologies in order to inform future practice, especially now that almost any technology under scrutiny will have some cyber-physical aspects to consider.
In this article, we conduct a systematic review of environmental assessments of CPS published in the peer-reviewed literature over the past decade, with the dual purpose of (a) describing the state of the literature and (b) distilling from the literature a set of system boundaries that researchers can reference to better capture the impacts of CPS. In the background section that follows we describe some key points from research on ICT and its environmental impacts in order to contextualize the relevance of routinely and rigorously applying a CPS framework in environmental assessment. For a thorough background on environmental impacts of ICT, the reader is referred to other articles such as Faucheux and Nicolaï (2011), Moyer and Hughes (2012), and Santarius et al (2020), whose qualitative studies are also included in our systematic review. We also refer the reader to conference proceedings where this work has especially evolved. The ICT4S Conference series, for example, was launched in 2013 and brings together researchers, government, and ICT industry practitioners to share insights on both the environmental impacts of expanding ICT and the potential for ICT-enabled systems to deliver reduced environmental impacts in other sectors (Hilty et al 2013, Chitchyan and Schien 2020).
1.1. Background: ICT and environmental impacts
Although the ICT industry consortium Global e-Sustainability Initiative (GeSI) estimated in 2015 that 'ICT has the potential to enable a 20% reduction of global CO2e emissions by 2030, holding emissions at 2015 levels' (GeSI 2015), greenhouse gas or other environmental benefits are not guaranteed, in part because ICT is a general purpose technology (Plepys 2002). Given that nearly every technology one might wish to evaluate now has some level of dependence on computing and communication, analysts must decide whether and how to measure the effects of this digital layer within a given system boundary, and to compare it to some realistic counterfactual (Amasawa et al 2018, Andrae 2018). Major challenges to adequately measuring ICT impacts are system complexity (Dahlbo et al 2013, Lago et al 2015, Whitehead et al 2015), the rapid pace of technological change and the continuous release of new, updated, and more powerful devices (Yao et al 2010, Cox et al 2018). Indeed, previous reviews have found that studies of ICT systems reach widely varying conclusions, with major reliability challenges associated with capturing rapidly changing system conditions (Yao et al 2010, Bull and Kozak 2014, Court and Sorrell 2020).
Environmental assessments aiming to inform policy and business decisions have taken different approaches to evaluating ICT systems. Beyond a focus on energy and climate, common across environmental assessment (Erdmann and Hilty 2010, Malmodin et al 2010, Court and Sorrell 2020, Masanet et al 2020, Mulrow and Derrible 2020), some have focused on the material life cycle of digital devices and network technology (Santarius et al 2020, Sovacool et al 2020). Because ICTs operate on extremely small-scale and precise performance, the life cycle impacts tend to concentrate in the upstream extraction, processing, and manufacturing phases (Huang et al 2009, Belkhir and Elmeligi 2018), in exchange for higher complexity and energy efficiency achieved during the use phase (Odum and Odum 2001). Recognizing the importance of economic growth and the sociopolitical context of digitalization for environmental impact, other analysts have evaluated ICT in macroeconomic contexts (Erdmann and Hilty 2010, Moyer and Hughes 2012, Salahuddin et al 2016, Santarius et al 2020). Still others call for a precautionary level of restraint on any kind of techno-optimism (Barry 2016, Cottey 2018, Grunwald 2018), recognizing the disconnect between optimistic expectations and actual outcomes (Kunkel and Matthess 2020, Lajoie-O'Malley et al 2020). Some of this disconnect is likely driven by second order effects, like rematerialization, induced consumption, and changed social practices that can lead to rebound (Börjesson Rivera et al 2014, Kern et al 2018, Bieser and Hilty 2020, Court and Sorrell 2020).
In addition to climate and materials there are health and socioeconomic concerns for global sustainability, and these call for further updates to life cycle assessment (LCA) and environmental assessment methodologies (Mulrow et al 2017, Grubert 2018). Three socioeconomic concerns are especially salient for ICT implementation: (a) the effects of automation on equity and the labor force (Cramer 2012, Mahmoudi and Levenda 2016, Murdock 2018); (b) the risk—especially in wealthy economies—that localized efficiency gains might mask a concomitant rise in impacts, environmental and social, in the distant places where materials are extracted and devices manufactured (Simas et al 2017); and (c) the 'creep' of new energy services added to the list of human wants and needs, as ICT enables new comforts and conveniences (Shove 2003, Mulrow et al 2018, 2019a). These claims and counter-claims about the potential impacts of ICT have received attention in the literature ever since digitalization trends emerged in industrial economies (Eason 1997, Jokinen et al 1998, Hilty 2008). Prior literature reviews have described indicators (Krumay and Brandtweiner 2016), methods (Kopelias et al 2020) and devised frameworks (Lago et al 2015) directed toward improving environmental assessment of ICT. Articles that have systematically reviewed methods related to environmental assessment of ICT include:
- Yi and Thomas (2007) summarized a variety of popular and peer-reviewed literature on the environmental effects of electronic business models and general ICT implementation. They identified many of the scope and scale questions that are still being discussed in the literature. Their review was presented in narrative format, and they only made a brief proposal for a machine-based method for capturing the variety of possible impacts arising from e-business/ICT.
- Arushanyan et al (2014) and Pohl et al (2019) both conducted systematic reviews of LCA studies of ICT products, services, and applications. Both studies discuss the challenge of going beyond direct impact assessment of equipment and energy use, and Pohl et al (2019) organize impact assessment categories into a perspectives-versus-effects matrix.
This study combines approaches from these prior reviews. Our search for literature was expansive, going beyond LCA and investigating any study claiming to assess environmental impacts. Given our interest in how mainstream environmental assessments treat CPS, we limit the search by focusing on peer-reviewed journal articles and we direct our interpretation by creating a framework for analyzing each paper and making bulk conclusions.
1.2. Research description
Through a systematic literature review, we investigate the hypothesis that current studies of the environmental impact of CPS largely ignore cyber-physical subsystems and dynamics.
The review is guided by two primary research questions:
- RQ1) In environmental assessments of CPS, what system boundaries are used for analysis of environmental impacts?
- RQ2) To what extent are CPS system boundaries identified in RQ1 applied across the current environmental assessment literature?
Both research questions are investigated using the same body of research articles, systemically searched and filtered from the literature. Furthermore, RQ2 references results from investigating RQ1. This step-wise approach is justified, as we aim to (a) summarize the literature both through narrative and categorization and (b) leverage this categorization to draw conclusions about the current state of the literature.
Ultimately, this review challenges environmental research in all sectors to rigorously consider the current and future landscape of CPS subsystems and dynamics that ultimately drive an array of sociotechnical and environmental changes. To support such consideration, we synthesize the peer-reviewed literature to present CPS analytical boundaries and approaches currently in use and recommend approaches for research going forward.
2. Methods
In order to address our research questions, we undertake five review steps described in this section. First, we (a) systematically collect our corpus. To address RQ1, on how environmental assessments categorize CPS, we (b) define CPS subsystems and dynamics, based on author judgment. To address RQ2, on the extent to which the environmental assessment literature evaluates CPS, we (c) code each article's capture of the nine subsystems/dynamics identified in response to RQ1, based on whether the aspect was assessed, discussed, not considered; and (d) perform cyber-consciousness scoring, using a simple scoring approach developed for this work. Finally, we (e) check for a relationship between cyber-consciousness score and climate impacts, among those articles that conducted a quantitative greenhouse gas analysis.
2.1. Article selection
Given this review's scope of peer-reviewed environmental assessments of CPS, we used a search and filtering procedure in Scopus to identify a corpus of English-language articles published in academic journals in or after 2010, as follows.
2.1.1. Initial search
Scopus was searched for peer-reviewed journal articles published in or after 2010 that included at least one environmental assessment term AND at least one CPS-related term in the title, abstract, or keywords. The set of search terms was expanded through three search iterations, as additional important terms were observed among search results. After three iterations, the quantity of new articles captured under the search criteria diminished greatly. We acknowledge that additional articles fitting our selection criteria may yet have been missed by this process, however we are confident in having captured a robust sample sufficient to address our research questions. The final list of search criteria applied was:
- Environmental assessment terms: environmental assessment, environmental impact, carbon footprint, environmental footprint, life cycle assessment, life cycle analysis (including also 'life-cycle' with a hyphen).
- CPS terms: cyber-physical, cyberphysical, smart, industry 4.0, internet of things, IoT, ICT, connected, app-based, mobile app, mobile application, sensors, WSN, digitization, digitalization, virtualization (including also '-isation' suffixes).
This search yielded 3127 articles.
2.1.2. Title and abstract filtering
The initial search was downscoped based on author review of titles, then abstracts to confirm articles were multicriteria environmental assessments of a CPS. The full corpus was reduced from 3127 to 320 articles based on title, and further to 146 based on abstracts. Common reasons articles were excluded include that they were: energy-only assessments (no environmental impacts evaluated), health impact assessments, assessments of solutions lacking a digital or ICT component, or descriptions of engineering innovations with only a brief discussion of emissions included.
2.1.3. Full read, filter, and categorization
Each of the 146 articles was then evaluated in full, resulting in three articles excluded due to inaccessibility. An additional 24 articles were excluded due to the analyses not fitting selection criteria beyond the abstracts, e.g. no environmental assessment, no analysis of cyber- or digital aspects, or insufficient information about system boundaries. The remaining 115 articles were then categorized as either quantitative (79) or qualitative (36) type articles, depending on the analytical approach and the type of results presented (see supplementary material S1 for further detail (available online at stacks.iop.org/ERL/17/013001/mmedia)). Qualitative articles included review articles and commentary pieces focused on the environmental impact of CPS-related technologies. In total, 115 articles were evaluated and included in our results.
2.2. Defining CPS subsystems and dynamics
Nine influential aspects of a CPS were defined through an iterative process of identifying and describing the analytical boundaries included among evaluated articles. Although the authors began this project with prior experience describing and researching CPSs, the extent of literature reviewed inevitably revealed new classifications and definitions that were useful in capturing the potential impacts of CPS. This process ultimately yielded five CPS subsystems and four CPS dynamics. Each CPS subsystem represents a set of component types that are similarly arranged within a CPS and that share environmental impact pathways. They are defined as:
- Energy system: fuels and other energy carriers required for system production/operation.
- Digital equipment: components directly involved in system operation and that are digital, e.g. on-site sensors, receivers, transmitters, servers, computers, tablets.
- Non-digital equipment: components directly involved in system operation and that are non-digital, e.g. on-site structures, mechanical components, lighting, books.
- Automation and management system: computational logic (as e.g. algorithms, programs, software) and supporting functions such as programming, development, and maintenance.
- Network infrastructure: components that support networked communication and storage functions, e.g. data centers, server farms, transmission and routing infrastructure.
CPS dynamics are outcomes of CPS implementation that are similar in scale and scope, and which have an influence on environmental outcomes. These are:
- Direct costs: net costs of implementation, for individual users and/or businesses.
- Social and health effects: impacts on behavior, equity, well-being, and human health.
- Impact feedbacks: second-order and/or rebound effects, e.g. direct or indirect rebound, induced consumption, rematerialization.
- Cybersecurity: impacts related to ensuring information security; physical, e.g. redundant storage or transmission pathways; or management structures, e.g. policies.
2.3. Article evaluation and coding
Once CPS subsystems and dynamics were defined, all 115 articles were revisited in order to evaluate whether the article included the nine categories within their analysis and/or discussion. For quantitative articles, if the article calculated environmental impacts associated with the system or dynamic, it was coded as having assessed (A) that specific category. If the system/dynamic was discussed as important or potentially influential but environmental impacts were not quantified, it was coded as having discussed (D) the category. For qualitative articles, if the system/dynamic was evaluated or described in detail as a driver of environmental impact, it was coded as an assessment (A); and if the system/dynamic was discussed as important or potentially influential but not directly evaluated under the review or study design, it was coded as a discussion (D). For both quantitative and qualitative articles, if the system/dynamic was not addressed as relating to environmental impact, it was coded as not included (N). All 115 articles received A, D, or N codings for each of the nine subsystems/dynamics.
2.4. Cyber-consciousness scoring
Subsequent to coding, each article received a summary 'cyber-consciousness' score, meant as a first-order measure of the extent to which an article considered the range of CPS concerns represented in the literature. An article receives two points for each A coding, one point for each D coding, and zero points for each N coding. The cyber-consciousness score is then the sum of all points received. The maximum possible points is 18 (nine categories × two pts each) and the minimum possible points is 2, since environmental impacts must have been evaluated for at least one category in order for the article to meet the review's selection criteria. An overview of article coding and cyber-consciousness scoring criteria is depicted in figure 1 and the method is further explained, along with examples of article coding conducted by the authors, in supplementary material S1. This guidance, along with the score sheet provided in supplementary material S2 (a Microsoft Excel file), are meant to assist in evaluating, implementing, or improving on our method.
Figure 1. CPS subsystems, dynamics, and cyber-consciousness scoring. A = assessed; D = discussed; N = not included in study.
Download figure:
Standard image High-resolution image2.5. Greenhouse gas results evaluation
Forty-three articles conducted a quantitative analysis of greenhouse gas emissions and compared results for a non-digital technology and a digital one performing the same or similar functions. For each of these articles, we calculate percent change in calculated greenhouse gas emissions between the digital technology evaluated and the non-digital technology to which it is compared. We then conducted an analysis of cyber-consciousness score versus greenhouse gas results across these 43 articles, testing for whether our formulation of cyber-consciousness shows a relationship with the sign or magnitude of greenhouse gas measurement.
3. Results
These results are presented in relation to our two research questions. First, we review subsystems and dynamics that environmental assessments may use to evaluate CPS technologies (RQ1). Along with each category description, we provide information on the extent to which articles in the evaluated literature assessed or discussed each system/dynamic (RQ2). To summarize this information, we present trends in cyber-consciousness score including an analysis of cyber-consciousness and greenhouse gas impact measurements.
3.1. CPS subsystems and dynamics
3.1.1. Energy system
Nearly all evaluated articles (97%) assessed or discussed energy resource requirements and their environmental impacts. Many of the evaluated articles focus on describing a new or emerging technology and then use fuel and emissions analysis to describe the technology's potential benefits. For example, Jeong et al (2015) describe an inter-vehicle communication system that would allow vehicles to lower speed and change lanes in response to surrounding vehicles' activity. They then compare emissions between equipped and non-equipped vehicles based solely on vehicle fuel use, measured via simulation. Makridis et al (2020) also evaluate emissions from AV technology in this way, describing their metric of focus as, 'change in the total fuel consumption (and consequently CO2 emissions that are directly proportional to fuel consumption).'
For many articles, the energy system is the only subsystem evaluated, and environmental impact is often derived by applying a greenhouse gas factor to the measured fuel or energy quantity. This is especially common for articles focused on buildings and transportation, where the digital intervention enables energy efficiencies within the building or on-board the vehicle (Ottelin et al 2015, Scheepens and Vogtländer 2018, Patella et al 2019, Noussan and Tagliapietra 2020). Meanwhile, energy use for equipment manufacture, software operation, and network infrastructure (other subsystems) is not included. This focus on the energy system outcomes of digital intervention is criticized in Murdock's (2018) article, which notes, 'These visions of 'weightlessness' push the routine labor of material production, maintenance, and disposal to the edge of attention.'
Because a CPS is inherently multi-nodal and geographically disperse in its functioning, environmental assessments must frame their analysis beyond energy systems, especially when the CPS is aimed at achieving localized fuel or energy savings.
3.1.2. Digital equipment
Digital equipment was the second-most addressed CPS subsystem among evaluated articles (66%). We observe that digital equipment is well-described in almost every article, even if the environmental impacts are not addressed. Many articles describe a CPS, such as AV controls or mobile app interactions, in technical detail, and then exclude equipment materials, production, management, and disposal from their analysis. Brazil and Caulfield (2013), Lee et al (2013), Yang et al (2016), Bento et al (2019), and Ko et al (2019), for example, all claim improved sustainability 'performance' or reduced 'environmental impacts' without including an evaluation or discussion of resource inputs to the necessary digital equipment. We highlight this as a critical gap in the literature, given the risks of further normalizing the 'weightlessness' of digitally enabled technologies (mentioned above).
We emphasize that material impacts of digital equipment could be evaluated given the content researchers generally included in their technology descriptions. Pipattanasomporn et al (2014), for example, evaluate a smart streetlighting system and include a detailed description of the hardware required—including computing and communication components, sensors, and a circuit diagram. Such detail lends itself to life cycle inventory and subsequent LCA, however, their environmental impact evaluation consists only of electricity usage multiplied by a greenhouse gas factor. Mutchek and Williams (2010) demonstrate a proxy method for evaluating material impacts. They estimate greenhouse gas emissions from the production of a smart irrigation controller by applying emissions data from the electronics manufacturing industry.
Several evaluated articles note the pace of change in digital technologies as an especially challenging aspect of making timely environmental assessments. Both Bull and Kozak (2014) and Yao et al (2010) credit Moore's law—describing steady growth in processing capacity since the 1970s—with creating such rapid technological updating that a clear picture of environmental impacts is difficult to determine through careful research. The same law is also used by authors to project future data processing efficiencies and hence improved environmental performance for AVs (Gawron et al 2019) and personal computing devices (Teehan and Kandlikar 2013). Finally, Bashroush's (2018) analysis of hardware refresh trends in data centers reveals that although Moore's Law shows the potential for reduced energy use, data center managers often refresh to the same level of power draw, using the efficiency gains to take on more business or provide a security cushion.
Thus, when setting up analytical boundaries for the impact of digital equipment, researchers should consider rates of equipment turnover.
3.1.3. Non-digital equipment
Non-digital equipment was assessed or discussed in 61% of articles. Whether or not this subsystem is included in an environmental assessment is greatly dependent on the baseline to which a given digital technology is compared. In the case of print versus digital media (Borggren et al 2011, Moberg et al 2011, Achachlouei and Moberg 2015), there is a definite need to consider the material makeup of magazines, books, newspapers, as well as their modes of production and transport. Other sectors may not have as clear a corollary, however many articles point to good reasons for considering non-digital equipment across technologies.
Gawron et al (2018) conduct an LCA comparing AV and non-AV systems. They include the vehicle body in their assessment since the addition of autonomous controls also changes certain non-digital aspects of the vehicle, most importantly weight and drag. This is an important example of how adding a digital component to a technology may necessitate design changes to non-digital components. Manosalvas-Paredes et al (2019) and Marmiroli et al (2019) also demonstrate the importance of including non-digital equipment within the scope of analysis. Their assessments of on-road wireless EV chargers and embedded pavement sensors (respectively) include the construction and maintenance of the roadway, whose design parameters and maintenance schedule are affected in the digitally enabled scenarios.
Based on these examples, we emphasize the importance of considering non-digital equipment in CPS assessments. It is worth considering what non-digital aspects are affected by the introduction of sensing, processing, and communication hardware on which CPS are built. In Aftab et al's (2020) study of an app for guiding drivers to open parking spots, they calculate the potential time and fuel savings given widespread adoption. What would be the effect on non-digital wayfinding infrastructure, such as signage and advertising, given widespread adoption? In the case of Li et al's (2016) assessment of sensor-based fertilizer application, what are the effects on fertilizing and farming equipment, given changing use patterns? These questions were not addressed, but represent the kind of additional impact information that could be gathered in CPS studies.
3.1.4. Automation and management system
Hardware, software, and resource demands develop together. As Whitehead et al (2014) point out, 'more powerful applications and faster speeds... [lead] to increases in the overall demand of data centres on power.' And yet the impact of developing and maintaining computational logic is included in the literature far less than those of energy and equipment (45% of articles). In Schmidt et al's (2018) description of a CPS for athletic stadium operation, they use 'automation layer' and 'management layer' to describe the control algorithm and setpoint configuration steps, respectively. We adopt these as a single CPS subsystem title because it captures a wider set of processes than might be included in the term 'software.' Jena et al (2020) describe the computational architecture of a CPS as consisting of: data acquisition, data cleansing and verification, data storage and management, analytics, and user access; noting that 'these layers are integrated and customized' to meet the purpose of a given CPS.
Thus, automation and management represent a tangible CPS subsystem that demands resource inputs and maintenance beyond those of hardware. In the case of Schmidt et al (2018), development of the initial stadium operation CPS was a three-year process involving a five-person research team. The study does not mention resources required for ongoing implementation, but does emphasize the CPS' ability to replace 'human-controlled' processing, hinting at a possible labor impact tradeoff. Zhang et al (2020) similarly estimate a high financial cost for the 'software layer' of their proposed blockchain technology; over 90% of the annual costs are spent on software development and maintenance. They do not measure the environmental impact of this layer, but do discuss it as potentially affecting environmental assessment of blockchain technologies.
Some articles include guidance on including automation and management impacts in environmental assessment. Malmodin et al (2010) include 'business activities' such as office operation and research and development in their greenhouse gas evaluation of the ICT sector. Kern et al (2015) specify a list of factors that affect the carbon footprint of software development: commuting, heating and ventilation, ICT infrastructure such as workstations and servers, and uninterrupted power supply. And Radonjič and Tompa (2018) detail a carbon footprint for telecommunications companies, including scope 1, 2, and 3 emissions.
3.1.5. Network infrastructure
Network infrastructure consists of the physical equipment enabling spatially dispersed communication and coordination. Specific structures include data centers, data transmission cables, towers, and satellites (Arts et al 2015, Lennerfors et al 2015). This subsystem is assessed or discussed in 41% of evaluated articles, and most of these are ICT-focused assessments for which the network equipment is the system under evaluation.
For many other assessments network infrastructure is not included at the point of intervention and is thus not included in the analysis. For example, Yuli et al (2019) provide a detailed description of the communication devices involved in a precision irrigation control system including sensors and controllers installed in the field, a router housed nearby, and the online systems that include data storage in the 'cloud' and a user interface and store. These online systems are, however, not included in their detailed LCA. Mutchek and Williams (2010) also describe a smart irrigation system, but only evaluates impacts of the controller, explicitly leaving 'information technology equipment' and 'off-site' sensors out of the study scope.
Steenhof et al (2012) discuss the difficulty of capturing impacts of distant data centers, especially in a cloud environment where processes may be carried out across a rapidly changing array of connected servers. Gawron et al (2018) point to a proxy solution, applying a general energy-use factor for a 4G mobile network of 1.25 MJ GB−1, 'which includes the power consumption of the base station, telecommunications networks, and data centers.' We find that this critical energy-use subsystem is left out of most other transportation-related assessments, especially those for AVs and smart charging where wireless communication is a central feature (Baumann et al 2019, Limb et al 2019).
Bull and Kozak (2014) provide a comprehensive argument for why network infrastructure must be considered in environmental assessment, pointing out that CPS-connected devices increasingly consist of small sensors and user interfaces. This comes at the cost of 'shifting the impacts of their footprint from their operation to their construction and running of the cloud to which they connect.' Without evaluating this subsystem, researchers will continue to focus on devices and miss the growing impacts of their enabling infrastructure.
3.1.6. Direct costs
Direct costs are a CPS dynamic encompassing the financial costs of system implementation. A variety of cost-category boundaries are drawn in the literature ranging from an individual's costs of direct technology usage (Nilsson et al 2017, Gawron et al 2018, Lu et al 2018, Scheepens and Vogtländer 2018), to a company's cost of deploying a technology (Mutchek and Williams 2010, Bauer et al 2018), as well as collective costs to society for technology deployment and maintenance (Cerdas et al 2017, Bashroush 2018, Limb et al 2019). This is an important dynamic to consider because it communicates a technology's likelihood of adoption among consumers, businesses, and governments (Oláh et al 2020). The direct costs dynamic was the most commonly addressed dynamic, being assessed or discussed in 57% of evaluated articles.
Direct costs tend to be measured on a net-cost basis, in which the cost savings from implementing a given technology are subtracted from the costs of producing or installing it. For example, Nilsson et al (2017) calculate the cost of installing and using a real-time energy price visualization meter, and found that energy bill savings more than offset the equipment costs. Mutchek and Williams (2010) calculate net savings based on water costs for users of smart irrigation controller and Gawron et al (2018) calculate net savings for AV controls, based on more efficient driving patterns.
Some studies include time savings in their cost assessment. Lu et al (2018), studying autonomous taxi impacts, justify this as a real direct cost, valued as a percent of average hourly wage. The presumption is that users of autonomous taxis could continue working and producing economic value while traveling in an AV. Aftab et al (2020) make a similar calculation in their assessment of a parking space locator app. Bieser and Hilty (2020) make a strong case for why time usage is an important aspect of technological impact assessment. In their proposed analytical framework, the 'activation of dead time' can lead to direct cost effects via increased economic productivity, but can also have other impact feedbacks when this time is used for activities such as media consumption and shopping. Direct cost assessment is thus an essential step in any impact assessment, but it must be supplemented with a consideration of non-cost and feedback dynamics.
3.1.7. Social and health effects
Many environmental impact studies take on a 'triple bottom line' framework, requiring economic, social, and environmental indicators in order to form a more holistic picture of a technology's 'sustainability' potential (Oláh et al 2020, Sartal et al 2020). Thus, social and health effects make up another important dynamic for researchers to consider. Among evaluated articles, 45% assessed or discussed social and health metrics or concerns.
The most common health effect metrics among evaluated articles were midpoint indicators calculated as part of an LCA. These indicators, such as particulate matter emissions and human toxicity, are calculated by applying standardized factors to the technology's life cycle inventory, using a method such as ReCiPe (Louis and Pongrácz 2017, Gu et al 2019, Gangolells et al 2020). The main health impact that was assessed outside of the LCA framework was automobile accidents, a common metric referenced in studies of AV technology. However, though lower accident rates were mentioned in almost every AV article, only Lee et al (2013), Olia et al (2016), and Guerrieri et al (2020) quantitatively assessed crash avoidance.
Treatment of social effects was much more diverse, but showed particular concern for the society-wide effects of digital technologies on labor (via automation of industrial and/or personal tasks; Mahmoudi and Levenda 2016, Murdock 2018, Jena et al 2020) and consumer behavior (especially via electronic shopping methods and advertising; Cerdas et al 2017, Briem et al 2019, Walzberg et al 2020). Unlike for health effects, there are no widely shared methodologies for assessing social effects among the evaluated articles. Christensen and Rommes (2019) analyzed interviews of young people about their use of ICT devices and delineated three social 'scripts' that characterize their understanding of electronics and environmental impact: (a) portable devices act as extensions of the body; (b) devices enable more rapid switching between tasks; and (c) devices are designed/programmed to be 'ubiquitous', constantly calling attention to themselves. These scripts ultimately helped explain why electronics users have trouble thinking about how to modify their use of electronics to reduce energy and environmental impacts. Identifying social scripts created, reinforced, or counteracted by a new technology could be a way to uncover both social and environmental effects of CPS.
Several articles mention concerns about the equity of digital solution implementation and control, but few evaluated or measured potential outcomes. Ottelin et al (2015) made an important observation that income level is more important in determining carbon footprint than any home energy efficiency strategy that might be implemented (including smart meters and automated energy management systems). We thus emphasize that checking the results of a technology evaluation against health and social factors provides a necessary context for holistic environmental evaluation.
3.1.8. Feedbacks
When impact assessments consider the five CPS subsystems they are typically focused on measuring direct environmental impacts that result from the production, use, or end-of-life stages of those systems. Many articles classify these direct impacts as 'first order' impacts, distinguishing them from 'second order' impacts that comprise a variety of system-scale effects of technology implementation (Erdmann and Hilty 2010, Faucheux and Nicolaï 2011, Bonvoisin et al 2012, Börjesson Rivera et al 2014, Das and Mao 2020). We give this dynamic a more generalized term, feedbacks.
Börjesson Rivera et al (2014) provide an exhaustive list of feedbacks that may be considered in environmental impact studies, including: direct/indirect economic rebound effects, time and space rebound effects, induced consumption, rematerialization, learning in production and consumption, and changed practices. Effects such as these were assessed or discussed in 43% of articles, more than social and health effects and more than the automation and management and network infrastructure subsystems. We take this as a strong signal that feedbacks are already an important and growing aspect of CPS environmental impact studies.
Rebound effects were an especially important concern for AV researchers, as it is widely recognized that innovations which reduce congestion on roads also tend to induce traffic increases over time (Bauer et al 2018, Gawron et al 2019, Liu et al 2019). A similar macroeconomic adjustment is noted for efficiency strategies such as smart homes (Ottelin et al 2015, Scheepens and Vogtländer 2018) and manufacturing (Kunkel and Matthess 2020, Santarius et al 2020). And most studies of digital media and communication technologies mention that rematerialization is a kind of rebound effect. This occurs when digital media displaces physical media but induces a new kind of physical material consumption in its place. Börjesson Rivera et al (2014) cite the example of digital music reducing purchases of compact discs (CDs) from the store, but also stimulating an entirely new habit of burning music onto CDs at home. The spread of additive manufacturing (i.e. 3D printing), and mass customization more generally, is raising similar concerns in the realm of consumer products (Cerdas et al 2017, Briem et al 2019, Kunkel and Matthess 2020).
In order to evaluate feedbacks, several articles utilize causal-loop diagramming, a system dynamics modeling technique. Erdmann and Hilty (2010) built a causal structure that connects the development and introduction of ICT technologies with the 'demand for the service' provided by that technology. Key factors affecting this demand include: speed of exploiting efficiency and substitution potentials, price per service unit, and price elasticity of demand. The authors can then model a variety of future demand scenarios, by setting informed values for each causal link. Bi et al (2020) similarly diagrammed economy-wide AV deployment based on causal links between technology design, cost, and macroeconomic demand. Finally, Moyer and Hughes (2012) model country-level carbon emissions from ICT as a whole. They link ICT deployment directly to changes in Gross Domestic Product (GDP), energy intensity, and cost of renewable energy, whose changes in turn affect overall energy consumption patterns and finally carbon emissions.
How society adjusts to a new technology may be hard to predict, but knowing that it must adjust—via factors such as behavior change, time-use, and cost dynamics—all environmental assessments should include some attempt at measuring or discussing feedbacks.
3.1.9. Cybersecurity
We categorize cybersecurity as its own CPS dynamic due to rapidly rising concerns about this topic throughout all economic sectors. This justifies giving it weight within the concept of cyber-consciousness, however we note that surprisingly few of the evaluated articles (8%) address cybersecurity at all.
In their discussion of ICT-enabled data collection for nature conservation, Arts et al (2015) identify data security as a factor that amplifies the resource requirements of ICTs. This tradeoff may or may not yield net environmental benefits as, 'more data and more analysis do not necessarily aid [conservation] decision-making.' They also mention the fact that ubiquitous sensors and cameras may pick up information on human activities, an issue that raises privacy concerns.
Thus, cybersecurity may be considered as a kind of feedback to environmental impact, as well as having social effects. Although the evaluated articles do not provide specific methods for measuring cybersecurity effects, analysts may look to methods from these other dynamics as they are increasingly called to address this concern.
Overall, the evaluated literature captured about half of the subsystem and dynamics categories. Table 1 provides article counts for each system/dynamic and the extent to which they were assessed or discussed.
Table 1. Subsystem and dynamics addressed within evaluated articles (115 total articles).
| Included in assessment (A) | Discussed but not assessed (D) | % of articles (A + D)/total | ||
|---|---|---|---|---|
| Subsystems | 1. Energy system | 110 | 3 | 98% |
| 2. Digital equipment | 60 | 14 | 64% | |
| 3. Non-digital equipment | 59 | 11 | 61% | |
| 4. Automation and mgmt system | 19 | 31 | 43% | |
| 5. Network infrastructure | 24 | 22 | 40% | |
| Dynamics | 6. Direct costs | 34 | 31 | 57% |
| 7. Social and health effects | 38 | 15 | 46% | |
| 8. Feedbacks | 21 | 29 | 43% | |
| 9. Cybersecurity | 1 | 8 | 8% | |
| Grand total | 366 | 164 | 51% | |
3.2. Cyber-consciousness scores
Cyber-consciousness scores clustered around the low and middle areas of the spectrum (see figure 2). Every single article with the lowest score of 2 shared a similar energy, fuel, and emissions-focused structure, assessing only the CPS energy system. The next cluster of articles, with scores between 7 and 10, represent a variety of approaches. Achachlouei and Moberg (2015), for example, included detailed evaluation of digital magazine readers, their enabling network infrastructure, their non-digital counterparts, and energy inputs to each of these aspects, receiving an 'A' (worth two points) in each corresponding subsystem category. Bonvoisin et al (2012), on the other hand, performed an energy and carbon analysis of wireless sensor networks based solely on direct inputs to the sensors and use phase energy. However, the article discussed software maintenance, network infrastructure, firm-level economic effects and larger-scale rebound effects, achieving eight points total. Only four articles achieved 15 or more points and all of these articles were qualitative assessments. Kunkel and Matthess (2020) analysis of ICT policy documents focused on the shortcomings of 'one-dimensional' policy expectations, namely that ICT-enabled efficiencies would deliver environmental and economic benefits simultaneously. In detailing why this is a shortcoming, the authors offer an assessment of nearly every CPS subsystem and dynamic. Murdock's (2018) description of digital systems' role in modern society details the many connections between the material realities of CPS and their social effects, thus touching on nearly every component except privacy and security. Cramer's (2012) article on human rights ramifications of ICT deployment similarly analyzed most CPS subsystems and dynamics. Finally, Bull and Kozak's (2014) systematic review of paper versus digital media LCAs scored highly because they use the disparities among LCA methods to discuss the complex array of physical and socioeconomic impacts of digitalized systems.
Figure 2. Cyber-consciousness score histograms; a) All articles, b) ICT sector, c) Transportation sector Blue: Quantitative articles; Gray: Qualitative Articles.
Download figure:
Standard image High-resolution imageThe evaluated corpus was dominated by articles from the ICT sector (analyzing digital systems generally, across consumer and industrial applications) and the transportation sector (with a major focus on traffic sensing and autonomous controls). These two sectors together made up 78 of the 115 evaluated articles. Their cyber-consciousness score histograms are also given in figure 2, showing a stark contrast in the type of articles published, and subsystem/dynamic coverage. Articles on the environmental impact of ICT are generally more qualitative, reflecting on current trends as well as a greater variety of subsystem/dynamic categories. While Transportation articles tend to be focused on specific technologies and largely conduct quantitative analyses capturing a smaller set of CPS subsystems and dynamics.
During the period studied, the number of environmental assessments of CPS has risen, from 6 in 2010 to 24 in 2020. The prevalence and diversity of CPSs has certainly increased during that time, and yet the average cyber-consciousness of articles do not show an increase. Figure 3 depicts the trend in article counts and scores over time, with average cyber-consciousness trending downward from 2010 to 2015 and moving between 6 and 8.5 thereafter.
Figure 3. Average cyber-consciousness (left axis); Evaluated article count (right axis), 2010-2020.
Download figure:
Standard image High-resolution image3.3. Greenhouse gas results
Among the 43 articles that included quantitative greenhouse gas measurement and comparison, the cyber-consciousness scores ranged from 2 to 14 (out of a maximum 18). Eight articles showed an increase in greenhouse gas emissions between the non-cyber technology and the baseline technology evaluated in the study, while the remaining 35 showed a decrease. A major outlier, showing a 431% increase, is the study by Bonilla‐Alicea et al (2020), which compared the emissions of riding a bicycle with those of an electric bike-sharing system.
Most studies show a potential decrease in emissions between non-cyber and cyber technologies, across cyber-consciousness scores. Figure 4 shows that overall there is not a significant trend indicating that a higher cyber-consciousness score affects the sign or magnitude of the projected difference in emissions from non-cyber to cyber technologies. This result provides important context to the cyber-consciousness concept as a scoping tool rather than a framework for affecting greenhouse gas footprinting specifically. Important background to this finding includes the low sample size of high-scoring papers as well as the fact that CPS impacts might be inherently more negative in emissions but positive in impact categories such as abiotic resource depletion.
Figure 4. Change in greenhouse gas emissions, non-digital/digital systems versus cyber-consciousness score showing interquartile limits and mean line.
Download figure:
Standard image High-resolution image
Article-by-article evaluation, scoring, and greenhouse gas result calculations are provided in supplementary material S2.
4. Discussion
The hypothesis that current studies of the environmental impact of CPS largely ignore cyber-physical subsystems and dynamics is broadly supported by the results of this literature review, with the major exception of energy systems analysis. The widespread inclusion of energy (assessed in 95% of articles; assessed or discussed in 97%) is not surprising, particularly given the historical (but changing) role of fossil energy demand as a proxy for environmental impact overall (Huijbregts et al 2010). The finding that other CPS considerations are inconsistently integrated is important to the field of environmental assessment because it means that our forecasts of technological impact do not provide enough context for decision-making with global scale and scope. Less than half of all articles assessed or discussed CPS dynamics including social and health effects and feedbacks such as the rebound effect. While acknowledging the difficulties and complexities of conducting holistic environmental assessments (Reap et al 2008a, 2008b), this analysis yet calls into question whether the environmental assessments currently being produced can effectively inform holistic environmental decisions about CPS technologies.
The results show that qualitative treatments of CPS impacts are well-aware of cyber system materiality and macro-level impacts, but that these are not well-modeled in quantitative studies. Researchers can tell, based on the known macro-effects of technological efficiency and the proven invisibilities of ICT infrastructure, that CPSs will carry environmental impacts that are both difficult to track and likely greater than typically predicted. However, quantitative analysts are quick to exclude such indirect effects from their analysis, possibly due to unfamiliarity with the phenomena or lack of available methods. That, or they are focused on mechanical efficiencies or direct use-phase inputs and effects, which can ignore wider system dynamics. Either way, it is likely that the preponderance of impact assessments are underestimates rather than overestimates of potential environmental impact. Due to the expansion of services that CPSs deliver, and the additional cognitive activities they imply, there are more ways for these systems to grow out of their initial savings than for them to contain consumption and secure lower and lasting aggregate impacts.
The literature also shows that analysts need not avoid CPS feedbacks, despite their more complex nature and nebulous boundaries. Modeling techniques such as system dynamics, as in Erdmann and Hilty (2010) and Moriarty and Honnery (2017), and system layering, as in Gawron et al (2018), have been successfully applied. Also the material structure of software systems, the internet, and data storage/transmission should not be a mystery. Its structure is well-described and modeled by many environmental assessments focused on digital systems, as in Bonvoisin et al (2014), Ipsen et al (2019), Marmiroli et al (2019), Das and Mao (2020).
We find the results of this review to be especially informative for researchers aiming to conduct environmental assessments as decision support tools (for policy, business, consumer decisions, etc). Looking at assessments from earlier in the decade, there is a wealth of data collected but it is questionable whether there was really a 'decision' to be made about the future of software distribution—on CDs or online (Williams and Tang 2012)?—or whether magazines would be read on digital devices or paper (Achachlouei et al 2015, Achachlouei and Moberg 2015). With internet connectivity on the rise generally, these product-based environmental assessments were not actionable, even had they shown the non-digital versions to be better for the environment. While some assessments may not have a decision-support purpose, environmental impact researchers are often called upon to interpret science for the purpose of making environmental decisions (Lubchenco 1998, Anderson and Bows 2012, Grubert 2018). In order to produce decision-relevant and timely research, environmental impact assessments should consider all CPS subsystems and dynamics during the scoping, question-formulation, and scenario development phase of research. Considering such a wide range of system boundaries can improve the decision relevancy and longevity of environmental impact studies.
Finally, we observe that macro-economic effects are especially troublesome for environmental modelers. It is difficult to predict how society as a whole will adjust to new modes of service delivery, time availability, and other indirect effects of CPS. But knowing that it does adjust on a grand scale means that macro-economic considerations can help contextualize environmental assessment. In table 2 we present some best practices for capturing CPS subsystems and dynamics from evaluated articles. We focus on those aspects beyond energy and digital/non-digital equipment, where boundary-setting and evaluation techniques were more diverse.
Table 2. Best practices for assessing CPS subsystems and dynamics.
| Automation and management | Network infrastructure | Direct costs | Social and health effects | Feedbacks | Cybersecurity |
|---|---|---|---|---|---|
|
|
|
|
|
|
Limitations of this work include the exclusion of non-English language articles, which prevents us from capturing the full range of assessment approaches being practiced around the world. We also acknowledge that the list of evaluated articles was refined in January 2021, meaning works published since then are not captured in this review. Our time series analysis shows that a widespread emergence of more cyber-consciousness methods and articles is unlikely to have occurred during this time.
We also acknowledge that many conceptualizations of both CPS and environmental impact that are possible. There is thus a certain subjectivity in defining subsystem categories on which to base an article score, such as the one we have presented here. We emphasize that this categorization was important for distilling and organizing our findings from a large body of literature, and we provide a high level of detail on article scores and the scoring method in supplementary material S1 and S2. Among potential evaluation categories, this review did not capture spatial boundaries, technology deployment scenarios, and counterfactuals that environmental assessments use to characterize digital technologies. These are parameters that also affect measurement and comparison among systems. In our ongoing research into EV charging systems, we will build on the CPS subsystems and dynamics defined in this paper and will especially consider spatial boundary and counterfactual configurations.
5. Conclusion
In this literature review we identified some clear signs that environmental assessment must continue to expand its scope of analysis to capture more CPS subsystems and dynamics. First, qualitative assessments generally have higher cyber-consciousness and have a much stronger focus on social and health effects and feedbacks, compared to quantitative assessments reviewed in this work. The qualitative literature makes a good case for why these effects are highly important to environmental assessment, and yet the quantitative literature lags behind in considering them. Next, we uncovered an array of existing techniques for capturing categories that are least covered in the quantitative literature, especially software and management, network infrastructure, and feedbacks. The fact that some researchers have successfully built analytical boundaries to capture these subsystems and dynamics means that there is existing work for other researchers to build on.
Finally, it is clear from current events—especially the rise of sensor networks, autonomous controls, blockchain technology, and cybersecurity incidents—that CPSs are the new norm across technological and infrastructural realms (Mora et al 2018, Lee et al 2019, Mulrow et al 2020, Pretyman et al 2021). In this light, it is alarming to find that the average cyber-consciousness of environmental assessments has been decreasing in recent years, at least for the criteria assessed in this review. We thus conclude the article by emphasizing that modern environmental impact assessment must consider CPS subsystems and dynamics during the scoping and question-formulation phase of research. This will improve the ability of techniques such as LCA and emissions footprinting to provide relevant and timely decision support. Here we note the emergence of assessment techniques such as dynamic LCA (Chopra et al 2019) and Futura (Joyce and Björklund 2021) that incorporate iterative system boundary updating and scenario development, respectively. Our ongoing work on EV systems will also incorporate the CPS subsystem and dynamics framework developed here.
In summary, this review investigated how the recent peer-reviewed literature treats environmental implications of CPS. We identified nine CPS subsystems and dynamics addressed in this literature and evaluated each article's analytical treatment of each. We find that current studies of the environmental impact of CPS technologies largely ignore cyber-physical subsystems and dynamics, except for energy systems evaluation that is widespread in environmental assessment.
Acknowledgments
This research was supported by a grant from the National Science Foundation Division of Computer and Network Systems (CPS Award No. 1931980). We are grateful to Tim Frick for valuable discussion and feedback.
Data availability statement
All data that support the findings of this study are included within the article (and any supplementary files).