Environmental Footprint Reporting as a Performance Metric in Environmental Data Science

Environmental data scientists often study the environmental impacts of human activity, but it is uncommon for scientists to quantify the impact of their own work on the environment. We performed a review of the literature published in the journal Environmental Data Science and coded papers based on whether they reported the environmental impacts of their work or the hardware information used in their study, which could be used to estimate the environmental impact of the computation. We find that only 1.9% of studies (3/155 assessed) consider the environmental footprint of their own methods, while 24.5% report their hardware specs. We advocate that environmental footprint should be reported and evaluated as a standard output alongside traditional outputs like R2 for computational tasks. To aid in this proposal, we developed and released a package, CodeCarbonR, that facilitates environmental footprint quantification of computational tasks in R. Adoption of this standard would allow the field of Environmental Data Science to create new norms around disclosure and transparency of the methods used and their impacts, which is especially important in the face of emerging technologies like AI.