Next GRASFI Annual Conference: 25 – 27 August 2025 in Paris. 

HomePaperUsing Graph Deep Learning to Trace Indirect Sourcing in Agricultural Supply Chains

Using Graph Deep Learning to Trace Indirect Sourcing in Agricultural Supply Chains

26 May 2023
Authors: Nataliya Tkachenko (University of Oxford), Juan Sabuco (University of Oxford), Christophe Christiaen (University of Oxford), Steven Reece (University of Oxford), Nicola Ranger (University of Oxford), Ben Caldecott (University of Oxford)
Presenter: Juan Sabuco (University of Oxford)
Abstract:

Corporate greenwashing is one of the most problematic areas in sustainable finance research, and is a well-known activity when companies position themselves as sustainable, yet fail to avoid reputational controversies suggesting otherwise. Such disputes can result from incomplete or conflicting disclosure reports, investigative journalistic inquiries by NGOs or internal corporate communication gaps, and hence are increasingly regarded as a data-related challenge, which has potential to be resolved with help of AI. One of such problems, known as indirect sourcing, is currently making headlines of many news articles as it reveals the massive scale of existing blind spots in evaluations of industrial impacts on nature and climate. The sectors of economy which make direct use of the primary commodities, such as heavy industry, pulp and paper and agriculture are under particular scrutiny as aside from the significant emissions budget they also make impacts on ecosystems, either directly adjacent or distant from their immediate regions of activity. Although supply chains are rarely designed to be transparent in the first place, the current best efforts are primarily undermined by the lack of relevant information, especially on the details of upstream supply chain actors. Using the case of beef industry we modelled spatial relationships between meat packing assets (abattoirs) and upstream cattle suppliers (ranchers), using fractal principles of the transport infrastructure arrangement. Our results demonstrate that AI can differentiate direct and indirect suppliers in the upstream chain of each individual meatpacking facility with the confidence 96.7%. This asset-level based approach is the first ever to demonstrate which meat-packing facilities are at risk of stranding due to configuration of their upstream sourcing catchments and exposure to indirect (bling-spot) spatial procurement shortcuts. Finally, we provide evidence that this classification approach can be easily translated to other industries, geographies and impact analyses, is both scale and distance agnostic and ethically aligned with the business information protection.

You May Also Like:

14 August 2024
Authors: Quyen Nguyen (University of Otago), Ivan Diaz-Rainey (Griffith University), Adam Kitto (University of New South Wales), Nicholas Pittman (EMMI), Renzhu Zhang (University of Otago)
14 August 2024
Login to your account