Beef, Banks and the Brazilian Amazon

Country/area: United Kingdom

Organisation: Global Witness

Organisation size: Big

Publication date: 2 Dec 2020

Credit: Chris Moye, Marco Mantovani, Louis Goddard

Project description:

This investigation reveals how Brazil’s biggest beef trading companies – JBS, Marfrig and Minerva – are linked to tens of thousands of hectares of illegal deforestation in the Amazon state of Pará. Despite relatively clean bills of health in private audits, our analysis shows how the firms’ suppliers – both direct and indirect – are deliberately flouting the law, with the companies themselves turning a blind eye. To conduct this analysis, Global Witness’s data investigations team had to build a picture of the beef traders’ supply chains using a range of publicly available datasets, automated techniques and manual research.

Impact reached:

The model Global Witness has developed is being adopted by other journalists and civil society organisations (CSOs), like Brazilian newspaper Folha de São Paulo, for similar investigations in other Amazon states. Previously, CSO investigations have focused on individual case studies or company sourcing in areas of high deforestation – whereas this methodology allows groups to definitively track direct connections between traders and deforestation. 

This work also contributed to long-standing civil society efforts calling for accountability from beef traders. On the day that JBS was due to reply to our allegations as we prepared our report it announced that it would begin monitoring its indirect suppliers by 2025 – an issue it had committed to address over a decade ago but had failed to act on. This data investigation showed the real-world consequences of international banks, asset managers and investors failing to insist on basic data that is needed to do due diligence on companies operating in high-risk sectors and jurisdictions. This investigation directly elevated discussions of how proposed legislation in the UK to tackle global deforestation through its supply chain – and similar measures in the EU – needed to cover financing. This included a new amendment on finance being tabled to the UK Environment Bill in the House of Commons. 

Investigators also held meetings with Brazilian Environmental Federal Prosecutors after publication that monitor the beef companies compliance with their commitments, and verbally told them they had demanded from the beef giants a response concerning our allegations. Finally, as a direct result of our work, JBS blocked a ranch featured in our report – El Shadai – after the company had purchased from the ranch in consecutive years despite it having been blacklisted by Ibama (Brazil’s forest inspection agency) for illegal deforestation. 

Techniques/technologies used:

To conduct the analysis, Global Witness’s data investigations team had to build a picture of the beef traders’ supply chains. We used documents called animal transit guides (GTAs), filed by ranches with the Pará state agriculture agency when they move animals around – for example, when selling cows for slaughter or export. The GTAs are available to consult one-by-one through an online portal. By guessing the documents’ ID numbers and automating the process through web scraping, we were able to obtain more than three million GTAs dating from 2014 to 2020.

Animal transit guides are just one piece of the puzzle. To quantify deforestation, we used a data set called PRODES, published by Brazil’s federal space agency, which shows areas of deforestation detected in satellite imagery. We then overlaid this data with information from the Pará state rural environmental register (CAR), which shows the boundaries of ranches, self-reported by their owners. The three data sources were combined automatically to throw up a series of suspect ranches, which were then investigated manually by Global Witness researchers, utilising landsat and sentinel imagery to remove, for example, any possible cases of false positives.

Our scraper was written in Python using the Scrapy framework. We initially experimented with doing the geospatial analysis in R, before switching to Postgres/PostGIS. Case studies were then investigated in detail using QGIS. Global Witness’s data investigations team provided a template QGIS project to help our expert forests investigators use the software.

What was the hardest part of this project?

This project presented a number of challenges. First, restrictions on access to the Pará GTA portal made it difficult to scrape, and we had to spend a significant amount of time and effort (tweaking scraper settings, renting servers in Brazil, etc.) developing an approach which would deliver consistent results.

Once the data had been collected, it needed to be processed and joined at scale with other publicly available data sets using fuzzy matching, requiring us to develop complex heuristics for determining accurate matches, particularly between cattle sales documents and farm ownership records.

Following on from the data collection and automatic processing steps, Global Witness’s forests team painstakingly verified deforestation in hundreds of individual candidate ranches using satellite imagery, working closely with our Brazilian partner organisation Imazon, building up a database of confirmed cases that could be used in the final report. Additionally, the data team had to collate deforestation permits and overlay these over the ranches to check on the legality of the forest clearance, in order to quantify the rate that was illegal. Ultimately, these were the most difficult and complicated parts of the project.

What can others learn from this project?

The project provides a template for data-driven investigation of agricultural supply chains for deforestation risk. The specific methodology set out in detail in our report and the code published on our GitHub page allows fellow journalists to recreate Global Witness’s entire research method, from scraping to automated analysis to manual verification using satellite imagery. But it also offers a starting point for expansion of this work to look at other forest-risk commodities such as soy and palm oil and other jurisdictions, both in Brazil and internationally. The modular approach taken means that new data sets can be slotted into the methodology relatively quickly and simply.
Beyond the technical methodology, the level of detail provided in our report on the editorial process – particularly our analysis and refutation of the big beef traders’ initial responses to our findings – should prove useful to journalists working on these companies in the future, giving them some idea of the arguments likely to be deployed to counter allegations of deforestation in their supply chains.

Project links: