Methodology and research perspectives

Environmental challenges related to the digital sector have been studied for almost 30 years now. After the acceleration of scientific publications on the subject since 2010 and their arrival in the media space since 2018, we have gone from marvels to worries and alarms. Yet, we still know so little about it. We are facing an ocean of uncertainty where environmental, geopolitical, technical, economic or legal currents mix, and within which some islands of knowledge are emerging. I used to explain that my research work consisted first of all in maintaining a topography of these seabeds, and secondly, in going to dive quickly where the topography indicates a particularly critical and little studied point. This is what led me for example in 2020 to study the claims of positive environmental impacts of digitization, or in 2021 to dive into the environmental footprint of semiconductor manufacturers in Taiwan with my colleagues at UC Leuven.

A shared observation

Recently, some challenging proposals on the future direction of our research field have been published. In 2020, the research community centered around TU Berlin and the initiative Digitalization for Sustainability (Tilman Santarius, Steffen Lange, etc.) published a report titled Digital Reset which advocates for a massive institutional and industrial redirection of the digital sector towards a new ecosystem centered on the concepts of sufficiency, resilience, sovereignty and equity. This report acknowledges that the promise of positive environmental impacts of digitalization is still elusive and, as a result, regulation of large digital companies and their business models is becoming increasingly necessary. In short, the authors argue that we cannot expect digitalization-as-usual to go, by default, in the direction of transition policies, contrary to what other European reports suggest, such as the twin transitions of the European Commission’s Joint Research Center. This report has the merit of proposing a more solid strategic orientation for European public policies if, and only if, the latter aim first and foremost to meet the objectives of ecological transition and social justice.

If the direction of public policymaking is changing, what about the scientific community? Two scientific papers published in February 2023 propose interesting findings and perspectives, at least for me. The first is the paper entitled The world wide web of carbon: Toward a relational footprinting of information and communications technology's climate impacts by Anne Pasek, Hunter Vaughan and Nicole Starosielski. These names may not be known by my colleagues in environmental science, but they are much more familiar to people working in Science & Technology Studies (STS) or media studies. Nicole Starosielski has written, among other things, an important book on the historical, environmental, and political issues of undersea cables (The Undersea Network, 2015) and has been working for many years with her colleagues on digital infrastructure. Here, Pasek et al. argue that the global environmental footprint of the digital sector will never be truly calculable for six reasons:

• access to industrial data;
• bottom-up and top-down assessments;
• definition of the scope of the digital sector;
• geographic averages;
• functional units;
• energy efficiency.

It is a clear summary of the obstacles we face today that I highly recommend reading. I personally agree with their observation and that’s why I stopped looking at global estimates two years ago. The authors propose a methodological shift to what they call a relational carbon footprint, which they say is “an empirical and strategic orientation toward demarcating particular relationships between elements—geographic, spatial, technical, and social—within a broad infrastructural network”¹. The examples given to illustrate such a method are unconvincing, but I seem to understand their intent, and I think I had a similar approach when my colleagues and I worked on the Taiwanese semiconductor industry. The point remains the same: defining a global environmental footprint of the digital sector is far from being a priority for our field of research (although some must continue to do so) and new methodological inputs are needed. We will come back to this last point later.

The second paper reviewed here is the one by Kelly Widdicks et al. which gathers contributions from the British multidisciplinary research consortium PARIS-DE (Design principles and responsible innovation for a sustainable Digital economy) but also from researchers well known in our field of research: Lorenz Hilty, Vlad Coroama, Steven Sorrell (especially for the rebound effects) or Simon Hinterholzer from the Borderstep Institute². While the paper by Pasek et al. deals with the direct effects of the digital sector, this one, entitled Systems thinking and efficiency under emissions constraints: Addressing rebound effects in digital innovation and policy, deals with indirect effects and thus completes our reflection. The authors note, not surprisingly, that the current context promotes a technosolutionist perspective based mainly on efficiency gains. They then call for serious consideration of the rebound effects of the digital sector, particularly because this integration is crucial in the development of public policies and their compatibility with climate commitments. As such, Widdicks et al. identify several challenges:

• They acknowledge that most of the communication strategies used up to now have failed, and that a new way of talking about rebound effects must be found;
• Secondly, rebound effects are very difficult to measure because, being of a varied nature, they open up an infinite number of mechanisms to be modelled. The authors call for a shift from complex calculation models to scenarios and for more empirical research to improve our knowledge of the subject;
• there is an obvious tension between reducing rebound effects and the economic and social consequences of such effects (employment, standard of living, etc.) and that this tension must be discussed in all transparency with the stakeholders (elected officials, etc.)³.

In the final analysis, they point out that working with political and economic leaders to take into account rebound effects means deciding on GHG emission constraints and their targeting. In the current context, this horizon remains far away.

These two papers suggest that we are reaching the end of the road with our current methodological choices. This does not mean that we do not continue to advance our knowledge: we have a much better view of the power consumption of 4G/5G access networks thanks to the paper by Golard et al ⁴ or on the environmental footprint of integrated circuits ⁵, or even on rebound effects like the recent papers on the impact of telecommuting on travel ⁶ or on domestic energy consumption ⁷. Yet, we need to find new ways to understand the environmental effects of digitalization and especially to define a better way to support the development of public policies on the subject.

Another scale: the territory defined by its transition targets

If digitalization is really at the service of the ecological transition, then we agree that the goal of all this research is to accompany cities, regions, or countries in achieving their transition goals. We must therefore confront them: by how much this city must reduce its transport emissions, by how much this region must reduce the final energy consumption of its residential sector, etc. This means that we cannot start with a digital technology studied in isolation, but rather with that same technology in relation to an environment whose scale and nature are determined by the administrative framework of the transition goals. For example, to see the transition goals (especially climate / energy) in French territories you can refer to the SNBC (National Low-Carbon Strategy) for the country, to the SRADDET for the regions, and to the SCoT (Territorial Coherence Scheme) and PCAET (Territorial Air and Climate Energy Plan) for the inter-municipal scale. Beyond the surge of acronyms, these strategic documents are designed to be coherent with each other and to articulate the objectives of the SNBC at each level, while adapting to the particularities of each territory. This initial methodological positioning has many advantages:

• it limits the scope of the study, characterizes a whole set of parameters (political, social, economic, etc.) and avoids any extrapolation outside this territory;
• it defines from the outset a baseline scenario and its expected evolution in 2030 and 2050 (the time scale of climate-energy strategies) without resorting to abstract projections;
• the lack of precision of environmental data from digital technologies is partly compensated by the access to open data of the studied territory;
• the biases of bottom-up or top-down assessments are less important at a smaller scale;
• the environmental analysis can be contextualized on the basis of the territory’s own infrastructure (energy, water, telecom, etc.) with its own characteristics (energy efficiency, carbon intensity, etc.) and can avoid, as much as possible, national or international averages;
• the direct (life cycle footprint) and indirect (efficiency, rebound effects, etc.) effects of digitalization can be modeled together without the common risk linked to extrapolation;
• this method goes beyond environmental analysis and includes the issue of land use planning;
• such an approach is much more adapted to the support of public policymaking.

Of course, every methodological choice also has its dark side, in this case :

• the results obtained are only valid in the territory studied and require a lot of work to build and glue the pieces together;
• there is always a problem of perimeter on the digital system studied (production versus consumption): do we include the data centers mobilized outside the territory, do we include the manufacturing phase, … ;
• the environmental impacts to be studied are dictated by the perimeter of the administrative documents and therefore present a certain deficiency: it is necessary to take into account, for example, the use of water, water and soil pollution, etc. (this can be taken from other documents such as prefectural decrees, but these are not intended to be long-term strategies);
• the administrative documents have a certain latency (the legislation is still too reactive and not very proactive, even on ecological planning) and it is necessary to complete the analysis with all the climatic and environmental risks identified in the scientific literature;
• the approach is environmental and does not yet integrate other types of impact (social, economic, etc.);
• structural effects such as the evolution of a sector or economic activities are less visible;
• not all administrative levels have open or easily accessible data;
• depending on the territory studied, it may be more or less sensitive to rapid technological change (dense urban areas versus rural areas, post-industrial countries versus so-called developing countries, etc.), which may profoundly modify the prospective analysis.

The territorial approach to the environmental effects of digitalization is not a silver bullet, but after two years of testing, it seems to me to be the most appropriate one to really support public policymaking. Of course, it is an institutional vision that is reinforced here and that only marginally takes into account the interests of traditional economic operators. However, I think that we need to strengthen this type of analysis to counterbalance similar exercises (GeSI, GSMA) that generally serve as sounding boards for industrial interests without ever taking into account concrete transition goals. Moreover, there is a healthy conceptual reversal in this approach: from an environmental point of view, digitalization cannot be seen as a universal phenomenon, most of the time its effects can only be characterized on a small scale.

Another logic: work more on the environmental consequences than on the attribution of impacts

If the territorial approach defines a new framework, other tools must also appear to complete the exercise. It seems to me that we need to go beyond the life cycle assessment (LCA) used with an attributional logic, the latter being ill-suited to digital services and insufficient to feed public policymaking (unless we want to relive the disaster on the email LCA). Today, consequential logic deserves a closer look. To illustrate the difference, I will use the example of the footprint of a plane trip that Olivier Corradi, the CEO of Electricity Maps, gives in a intervention that I strongly recommend :

• attributional logic: I divide the emissions of the plane by the number of passengers;
• short-term consequential logic: the plane will leave without me whether the seat is occupied by me or not, my footprint is close to zero;
• Long-term consequential logic: if enough people decide to stop flying then a plane will stay on the ground and my decision will represent a small contribution, I divide the emissions of the plane by the critical mass necessary to keep it on the ground (here I avoid emissions).

Here the three methods of calculation are correct but they do not fulfil the same objectives. If we are looking for ease of calculation, attributional logic is obviously the simplest, unlike consequential approaches, which involve a complexity of calculation and a mixture of qualitative and quantitative data. If one is looking to send the right signal to enable behavioral change, the attributional and long-term consequential approaches are the most relevant, both indicating that less air travel is conducive to reducing emissions from the sector. If we are looking for the method that is closest to the physical phenomenon, then the short-term consequential logic is the most appropriate, but it sends the worst signal to change behavior.

It is easy to understand that attributional logic is more like the classic accounting exercise that takes place a posteriori of the action observed in a given time frame (the direct emissions of year n). Long-term consequential logic is rather a prospective exercise that consists in determining the present and future environmental consequences of an action (related to a service, product, system, etc.) in relation to a counterfactual scenario where this action would not have taken place. It requires the definition of at least three scenarios (the baseline scenario, the future scenario with the modeled action, the counterfactual future scenario without the modeled action).

Attributional logic is easier for the digital sector because we study widely distributed and shared systems and avoid asking ourselves what is the exact share of computing dedicated to such service in such machine. This logic also works well with the sector because we are also faced with black boxes that are rarely opened by economic actors, so in the absence of data we make allocations. However, a consequential vision naturally fits into the framework of analysis and public policy making. In fact, political discourse is saturated by consequentialist logics (often short-term and highly biased). For example, if an elected official says that the installation of a data center will create jobs in his city, to verify his hypothesis it is necessary to formulate a counterfactual scenario in which another industry potentially interested in the land could have installed its activities. One can then compare which decision creates more jobs. Beyond the traditional demogogic drifts, thinking from a consequential logic means asking where we want to take our society in the long term, and this is the political issue at stake in all the research we are conducting.

In fact, the consequential logic is already used in modeling the environmental impacts of digitalization, especially for indirect effects. There are mixed models where a consequential logic is used to identify the possible effects of a digital service, and the results of such effects are then attributed among the actors involved in the deployment of the service. Let’s take the example of a standard consequential tree for online commerce.

This scheme mixes short and long term consequences and implies a mixed accounting (attributional and consequential) to qualify and distribute the effects on emissions to the different actors of the studied solution. Generally, attributional accounting will be used to model direct effects (life cycle impacts) and consequential accounting to model indirect effects (efficiency, rebound effects, etc.). However, when we mix the two sets in the same assessment, we create double standards, especially when this whole structure is mobilized to calculate only avoided emissions. It is also worth noting that a consequence tree is always incomplete by nature, and here we can regret that the tree does not identify the increase in logistics and warehouse space to cope with the increase in online commerce. If not complete, a consequence tree can be applied to a specific territory to ensure that the priority effects in this context are well identified. If we had done this exercise with the City of Paris, it is likely that the development of dark stores as an effect would not have been forgotten⁸.

Today, this mixture does not inspire me much confidence and I tend to separate these logics according to their objectives. The consequential logic should not be considered as an accounting method (and not become a new mathematical modeling challenge) but as a tool to help decision-making, in particular to orient investment and the regulatory effort of public authorities. The attributional logic has rather a control function to ensure that the path traced by the consequential produces the expected results. It seems to me that the territorial approach with a consequential logic will make it possible to create decision support models that are much more relevant than what we have today.

	Attributionnal	Consequential (long-term)
Purpose	Define the distribution of impacts of past actions	Define the consequences of present and future actions
Time perspective	A posteriori	Next 10 years and more
Scale	At the scale of the system studied	Consequences identified but necessarily incomplete
Modeling	Classic environmental accounting	Scenarization
Link with policymaking	Inform on the effects of past decisions	Define compatibility with transition policies and help to target investments and guide regulation

Other scenarization methods

To define the environmental effects of a digital service you need to define a baseline scenario from which you will compare at least one scenario where the digital service is deployed. The difference between this baseline scenario and the scenario with the deployed service allows you to define if there are avoided or added emissions due to the service in question. As such, the definition of the baseline scenario is critical: a pessimistic baseline will inflate the potential avoided emissions and vice versa. As mentioned above, a counterfactual scenario should normally be modeled to estimate what would have happened without the introduction of the digital service. Counterfactual scenarios are usually hypothetical because once the service is deployed it is no longer possible to know what would have happened without it. If we project into the future, we must then scenarioize this baseline and its counterfactual evolution as explained below.

There are many problems with scenarization, but there is one that is of particular interest here. In the figure above, we see that the counterfactual scenario is to continue the trend in emissions before the digital service is deployed. Considering that emissions from activities will increase without the integration of a digital service is a huge bias that can make the whole analysis invalid. In fact, ecological transition policies would tend to reduce the baseline and its counterfactual evolution over time, this seems to be true in most cases except for some digital services. To tell the truth, it is while doing territorial analyses that I realized this obvious fact. Since my analyses are part of the territory’s transition objectives, then my baseline for 2030 and 2050 is already provided, and it’s decreasing. For example, if telecommuting avoids GHG emissions by eliminating a trip by combustion engine car, in ten years telecommuting will avoid a trip by electric car (if electrification) and/or by public transit (if modal share shift). Thus the emissions avoided can only be reduced in this case. I have discussed this with many colleagues in France and in Europe and we are almost all in agreement on this observation.

The corollary to this observation is that if the baseline decreases then the emissions avoided or added also decrease over time. However, most evaluations of the emissions avoided by digitalization consider that their gains remain stable over time. So if my baseline is decreasing over time then the methodological problem is to know what role the digital service plays in this decrease. For example, is the reduction in my home’s final energy consumption linked to insulation work, to an energy crisis leading to higher electricity costs, to a decrease in heating time due to milder winters or to my smart thermostat/meter? Again, there is a conceptual gap in the current methods that rarely take into account this type of external factors (public policies, crises, etc.). However, I am less interested in this point, even if it is necessary, because it again brings in complex calculations to attribute gains.

Similarly, is a change in behavior enabled by a digital service lasting? This is another important bias in current modeling. It is assumed that an effect, whether it is related to the immediate effectiveness of the new service or to a change in behavior, is constant over time, but there is no evidence that such a trend can be applied to all digital services in all places and at all times. In the absence of being able to follow the evolution of these effects over several years, I tend to take conservative but stable values over time to reduce the risks of overestimation.

Conclusion

My research perspectives are now clear to me. My research activities must serve the development of more sustainable public policies, specifically on the issues of digitalization. To do this, I am immediately in line with the transition goals of the territories in which I model a service - the right administrative scale today for me is the region. Thanks to this focus I can define baselines that integrate the effects of public policies and follow a consequential logic for the territory and for the digital service I am evaluating. The combination of the territorial scale, the consequential logic and the scenarization of transition policies aims at producing decision support models for public institutions that are submitted at the same time to the injunctions of ecological transition and digitalization. These models are not primarily intended to provide precise figures, but to identify the conditions for success and failure of a digital service so that it can help achieve the transition goals of a given territory. Of course, this does not solve everything, but in any case this positioning allows us to bypass many of the obstacles identified by Pasek et al. and Widdicks et al.

In fine, I direct my research towards a much more multidisciplinary approach and I aim at producing three types of analysis: a technical one, an environmental one and a political (or geopolitical) one. The paper on the environmental footprint of semiconductor manufacturers in Taiwan remains for me an archetype of the work I want to develop. We were able to produce a 6-year environmental analysis based on detailed knowledge of IC manufacturing, which identified a clash between industrial development and the island’s transition goals, leading to a carbon lock-in. Producing this type of analysis is of course only a first step as concrete actions must then be generated, but this will be the subject of another research cycle.

	New approach
Purpose	Enable public institutions to define whether a digital service helps meet transition goals
Scale	Territorial analysis defined by its transition goals (PCAET, SRADDET, etc.)
Modeling	attributional a posteriori / consequential for indirect effects and projections / mixed
Baseline	Monitoring of transition goals (assumption of success)
Result	Define the conditions for success and failure of a digital service to help achieve the transition objectives of a given territory