Scientific CO2nduct

A Framework for Reporting the Carbon Cost of Scientific Research

TLDR: Climate change and global heating poses an immediate and serious threat to us all. Transparency with respect to the effects of our actions is a key first step towards finding solutions. The goal of this project is to enhance the transparency of the climate costs of scientific research, through the use of a standardised reporting table which can be included as an appendix to scientific publications. Additionally, we aim to collect and develop additional ideas for sustainable science. We look forward to your contribution

The recent Intergovernmental Panel on Climate Change (IPCC) special report on the consequences of global warming to 1.5° C above pre-industrial levels has made it clear that global heating poses significant and urgent threats to the sustainability of both humanity and our planet. In order to avert global catastrophe and mass extinctions it is critical that all members of society acknowledge the scale and urgency of the crisis that we face, and act immediately both to address the sustainability of our own lives and to demand meaningful responses from government and industry. We believe that the scientific community has a particular obligation, not only to contribute towards education, advocacy and effective solutions, but also to be transparent in the evaluation and mitigation of the carbon costs of our own research. To this end, we propose here a simple framework for reporting the carbon costs of research, with the hope that both individual researchers and institutions will commit to such reporting. In particular, it is hoped that in the short term such a standardised reporting framework will raise awareness of the scale of our collective impact, and promote the immediate offsetting of carbon costs, while in the longer term we hope to inspire change in the ways in which global collaborative scientific research is performed, using sustainability as a design principle.

A Standardised Reporting Table

We suggest the inclusion of a CO2 reporting table, such as the one below, in all scientific publications. This table reports in a transparent way the carbon cost of a particular research work, such as a publication, taking into account the cost of both computational resources and research associated travel. We stress that the table below is simply an example, and we encourage the inclusions of any other CO2 costs which may be relevant to a particular research work (for example, electricity use required for laboratory equipment). The goal of such reporting tables is to make the carbon costs of research transparent, thereby raising awareness of this critical issue, and facilitating the development of a meaningful emission reduction strategy, for example along the guidelines of the science based target initiative .

Numerical Simulations
Total Kernel Hours [h] 14260
Thermal Design Power Per Kernel [W] 5.75
Total Energy Consumption Simulations [kWh] 82
Average Emission Of CO2 In Germany [kg/kWh] 0.56
Total CO2-Emission For Numerical Simulations [kg] 45
Were The Emissions Offset? Yes
Total CO2 Emission for Transport [kg] 1050
Were The Emissions Offset? Yes
Total CO2 Emission [kg] 1095

In order to facilitate the use of such tables, we have included a LaTeX template in the resources folder of the website repository. If you are not using LaTeX, feel free to use your own template containing the above information. Moreover, the table can be shortened if no numerical simulations were performed. If you decide to visualise your CO2 emissions, please also reference this website (see bibtex entry at the bottom of the page) to increase the visibility of this project among your community. You can also send us a link to your paper to be included in the list of examples which can be found below. Before getting to the details, we would like to emphasise a related idea for calculating the carbon footprint of machine learning simulations that might already give you all the relevant information for your calculations. Let's now walk through an example of how to calculate all the necessary information. We begin with a description of how to calculate the total CO2 cost of any computational resources which were utilised: Firstly, it is necessary to estimate the total kernel hours that were used. This should be available to users of High Performance Computing (HPC) systems, or can be kept track of manually. Next, one should estimate the thermal design power per kernel. For this, you will need to find out on which kernel the simulations were conducted. The thermal design power should then be available in the manual of the manufacturer. For users of HPC systems we encourage you to reach out to your system administrators to obtain this information. Given the above information, one can calculate the total energy consumption (in kWh) via

[total energy consumption (kWh)] = [total kernel hours (h)] x [thermal design power per kernel (W)] / 1000

Next, it is necessary to find out the average CO2 emissions (in kg/kWh) from energy production in the country in which your computational resources were located. For European countries this information, sourced from European Environmental Agency, can be found here. If you have additional data for your location, please add it to this website via a pull request! If you cannot find this information about your location or are unsure where the computations were conducted, we suggest using the global average of 0.47 kg/kWh. Finally, to calculate the total CO2 emissions (kg) of your computational resources one uses the simple formula:

[total CO2 emissions (kg)] = [total energy consumption (kWh)] x [average CO2 emissions (kg/kWh)]

We would like to emphasise that this calculation necessarily only yields a lower bound to the actual emissions as it ignores factors such as cooling altogether. Nevertheless, we think the value of this oversimplified estimation is to deliver an intuitive feeling of the involved climate costs. Any improvements to our methodology are of course warmly welcome. Moreover, we strongly encourage experimentalists to contribute a reasonable framework to estimate the energy needed for research conducted in laboratories as well. Before discussing ways in which the CO2 emissions from numerical simulations can be offset, let us first describe how to estimate the CO2 emissions from research related air-travel. The first step in this regard is to list all research related flights, and we would like to suggest including flights for collaborative research visits, as well as conference attendance in fields relevant to the research work, or at which collaborators are present. Given this list of flights, one can then use atmosfair to calculate the CO2 emissions of each flight, and sum them together to obtain the total, which is then reported in the table. Finally, at this stage we can discuss the ways in which the relevant CO2 emissions can be offset. Once again however we want to stress that offsetting emissions is not a sustainable solution, and provides only a temporary strategy, which we advocate in conjunction with dedicated efforts to re-design our research process in a way that drastically reduces emissions. Again, we also acknowledge that offsetting costs may be prohibitive for individual researchers, and we hope to provide motivation to students and researchers to insist that such costs be covered by funding organisations and academic institutions. Even if it is not financially possible to offset the relevant emissions, we still encourage using the reporting table, as a method of raising awareness to this issue. In light of these caveats however, given an estimate of the total CO2 emissions, if financially possible these can be easily offset using atmosfair.

Towards Sustainable Science

Given that emission offsetting is not a sustainable or effective approach to mitigating the effects of global heating, what actions can we take, individually or as a community, to move towards a more sustainable mode of scientific research? Here we provide some simple ideas, not all new, for how this might be achieved. However, we strongly intend this as a living document, and so if you have ideas or suggestions, please feel free to make a pull request! Let's begin with travel related CO2 costs:

Virtual conferences. Conference related air-travel is certainly one of the greatest contributions to the overall CO2 footprint of a researcher. While we recognise that such meetings play both a social role and an academic role, we strongly believe that in many instances virtual conferences allow for effective dissemination of research, with drastically reduced CO2 emissions. We also believe there is significant scope for creativity in format here - for instance a conference as a collection of pre-recorded talks, supplemented with scheduled online question and answer sessions. This in fact may even aid the dissemination of research, by allowing access to a potentially much broader audience than those who could attend physically, while maintaining the quality control of accepted talks.

Bundling conferences. As mentioned above, we acknowledge the essential social role played by conferences, in building community cohesion and facilitating the development of collaborative relationships and new opportunities, and we recognise that virtual conferences cannot fully satisfactorily fulfill this role. However, we suggest that bundling conferences - i.e. holding one much longer event per year, rather than many smaller conferences - would allow for effective community building, while again significantly reducing the number of flights an individual researcher is required to take per year. In many communities there already exist such events - for example the annual three week Benasque quantum information meeting (popular article in German and a translation in English).

Choice of location. If a conference is really necessary, choosing the location in a way that ensures many participants have the opportunity to reach it without the need for flying can make a big contribution towards mitigation of CO2 expenses.

Using green venues. By now, there are many venues which offer green conferences, extending from the power supply of the venue to providing food and drinks from regional resources. Additionally, organizers of conferences could make vegetarian meals the default option, with fish or meat needing to be requested specially. In a similar vein, for catered meals one could move towards displaying transparently the CO2 costs of different menu options. Ideally we should be both scientists, and ethical consumers.

While air travel is certainly the major contributing factor for research related CO2 emissions, numerical simulations and the associated computational resources are also significant contributors, particularly in fields such as machine learning and data science. To this end we make some suggestions as to how such emissions may be mitigated.

CO2 emissions as a design principle. When designing numerical simulations, scientists usually do not take CO2 emissions into account. The runtime and the availability of computational resources are usually the determining factors to choose the parameters for the simulation. We suggest a change in state of mind, allowing for questions such as "Will my result improve in a significant way if I emit more CO2?" , or, "Can I design methods and techniques with the goal of achieving the same results with less emissions?".

Visibility of CO2 emissions. One of the primary motivations for the reporting framework proposed here is precisely to facilitate the process of using CO2 emissions as a design principle, by making the amount of emissions visible. However we believe it is possible to go further by developing open source tools, and designing computational frameworks, that make visible the amount of CO2 which will be emitted either at the time of job submission, or in the user statistics for a particular computing facility or device. The widely used SLURM scheduler actually comes with such a feature!

An open source policy towards CO2 heavy results. It is becoming more and more popular to share code that is used for the scientific simulations. Large open-source libraries with special tools exists for almost any research direction. We propose to extend this sharing mentality to also share the results of costly computations, such as trained neural network models, or the outcome of other optimisations. While open source code makes running the model again easy, its not free for our planet.

A Concluding Remark

Our planet, and life as we know it, is in an urgent state of crisis. We believe that as scientists we have a particular obligation to acknowledge the scale and urgency of our situation, and to act in whichever ways we can to contribute towards the sustainability of our environment, both for ourselves and future generations. In order to raise awareness on these issues, and to stimulate both debate and change, we commit to transparently reporting the carbon costs of our research, offsetting these costs as a temporary strategy, and working actively in the long-term towards a sustainable mode of scientific production. In particular it is necessary that we all contribute to redesigning our own scientific process, and pressuring our funding agencies and institutions to support us in this effort. Our success relies on action as a community, and so we encourage you to make similar public commitments, and look forward to working together for the collective future of our planet.


The following papers include examples of carbon cost reporting tables, as per the guidelines above. If you include such a table in one of your papers and would like it included in this list as a show of support for this project, please let us know, or simply fork the repo of the website, add your paper, and make a pull request.

Cite as

If you include a carbon cost reporting table and would like to cite this website for context, or for any other reason, please use the following bibtex entry:

  title = {Scientific CO2nduct},
  subtitle = {Raising awareness for the climate impact of science},
  howpublished = {online},
  url = {},