Project summary

This is a joint proposal by a Chilean advocate, educator, and researcher, and a US-based scientist.

Factory farming is expanding into low- and middle-income countries, yet basic information including the numbers and locations of concentrated animal feeding operations (CAFOs) is surprisingly scarce. With that information, researchers could gather location-specific data about the impacts of CAFOs (e.g. monitoring water contamination, modeling disease transmission), and test hypotheses about where and why factory farming is proliferating. In turn, advocates would be better armed with evidence and information that would help them develop strategies and propose policies that could limit the expansion of industrial animal agriculture.

Neural networks have been used to identify CAFOs in the USA, so they could in principle be employed to detect CAFOs globally. Verifying neural network-based CAFO candidates would be greatly assisted by complete, accurate, geo-tagged lists of CAFOs outside the US. Here we propose to create such a dataset in three diverse regions of Chile. We will use the database in a larger project to eventually map CAFOs worldwide, and also to support farmed animal advocacy efforts in Chile.

What are this project's goals and how will you achieve them?

Our primary goal is to create a database of chicken and pig CAFOs in 3 regions of Chile. To the best of our knowledge, no such dataset currently exists for a country outside the Global North. Our secondary goal is therefore to understand the techniques and obstacles involved in creating one, and provide lessons that can be applied in other locations. We will work in Chile partly because it is a middle-income country with highly diverse landscapes (desert north to forested south), and partly because one of us (CL) is already deeply involved in, and committed to, understanding and resisting the development of factory farming there.

This proposal describes the wider CAFO-AI project in more detail. The main unknown is how variable large-scale, industrial CAFOs are in countries outside the US, and whether a model trained on US data would generalize well to other locations. We will begin to address that question with this project (REM is also in touch with animal advocacy organizations in Asia and Africa regarding potential similar work).

The current project’s methodology:

1. Search the records of the Chilean Agricultural and Livestock Service for livestock establishments (these records are public). 

2. Use Chilean transparency laws to obtain addresses and animal records.

3. Use satellite imagery to sort CAFOs from other farm types, and assign coordinates

4. If necessary, also:

   1. Search websites of the entities obtained in Step 1; contact them via email or phone.

   2. Contact local NGOs, journalists, and university researchers who may have knowledge about CAFOs 

   3. Look for relevant information at the Ministries of Health and Environment

We will collect this information for each facility: Name/ID, Coordinates, Animal Type, Size or Capacity, Parent Company (if possible). We will also produce a report that describes the methodology, presents satellite images and a brief analysis of CAFO characteristics, and gives lessons learned.

Paths to impact: 

1. Share report on Critical Animal Studies platforms & with Chilean NGOs→ Chilean advocates gain comprehensive view of number and impact of CAFOs → Propose evidence-based policies and use information for public outreach → Farmed animal suffering prevented, local animal rights movement strengthened.

2. Use data to train and test model to identify CAFOs in Chile  → Researchers better understand spread of CAFOs and their impacts → Advocates can create better strategies for resistance, intervention and education→ Farmed animal suffering prevented.
   

How will this funding be used?

REM: Salary and overheads to cover ~100 hours of work (project management/communications, data validation/curation, analyzing CAFO characteristics, writing report) @ $75/hr → $7500 USD + 20% buffer = $9000.

CL: Salary and overheads to cover ~250 hours of work (obtain and verify the CAFO information) @ $35/hr → 8,750 USD + 20% buffer = $10,500 USD.

Hourly rates account for health/dental/disability insurance, retirement savings, self-employment taxes and are appropriate for our respective countries.

Who is on your team and what's your track record on similar projects?

Carlos Liebsch will lead the data collection and local dissemination of results. Mr Liebsch is an activist, educator, sociologist and researcher in Critical Animal Studies, currently serving as project coordinator at Clafira Sanctuary. He has published research into the industrialization of animal agriculture in Chile and the Chilean fishing and aquaculture industry, and has recently received funding from VegTrust to continue these investigations. Furthermore, he teaches a university course on animal liberation, which involves the presentation of the research results.

Rachel Mason will ensure that the database is suitable for training and testing a neural network, verifying that each facility is tagged with accurate coordinates and suitable meta-data. She will also co-author the accompanying report. Dr Mason has previously trained a neural network to identify buildings in remote sensing data, so is qualified to lead the next phase of the project. As a mid-career scientist she has published numerous scientific papers and reports.

What are the most likely causes and outcomes if this project fails? (premortem)

The biggest risk is probably that Chilean government agencies could be unresponsive and/or provide vague or inaccurate information. They are legally obligated to comply with requests for the information we need, but that does not necessarily mean that acquiring the information will be easy, or that there will not be errors in the data. Whether or not we could work around this by using secondary data sources (e.g. using satellite images and personal contacts to correct erroneous data) would depend on the nature of the problem. At worst, it could make it impossible to achieve the primary goal we stated above. In that case, we would probably write a report, editorial, or blog post highlighting the lack of reliable CAFO data worldwide, why it is needed, and the difficulties we encountered in obtaining it.

What other funding are you or your project getting?

REM has been awarded $30K from the Food Systems Research Fund for a pilot project to develop a method for AI-based CAFO detection in LMICs. The current proposal is to supply the training and test data to be used in that work. (In principle, training a model on US data and applying it elsewhere might be possible with human-in-the-loop type methods, but dedicated non-US training and test data would be a big asset).

REM has also applied for a Smithsonian Senior Fellowship (the proposal linked above) to continue this work and connect with scientists who are interested in using a global CAFO database to study the impacts of industrial animal agriculture. 

We previously applied for funding for this project from the EA Animal Welfare Fund. They responded that the project was above their usual bar for funding, but could not be funded because of their current funding shortfall.