by Vanessa Meadu

New data on global livestock distribution to help sector along sustainable pathways

Today, the United Nations Food and Agriculture Organization (FAO), in collaboration with the Université Libre de Bruxelles (ULB) and other partners, releases a new, 10 km global dataset of livestock distributions: the Gridded Livestock of the World, version 3.0 (GLW3), published in Nature Scientific Data.

GLW3 has a reference year of 2010 and includes global distributions of cattle, buffaloes, sheep, goats, horses, pigs, chickens and ducks at a spatial resolution of 5 minutes of arc.

The digital maps in geotiff format are freely available for download via FAO’s Livestock Systems website, which also provides data on production systems and links to resources related to sustainable livestock sector development.

Timothy Robinson, Livestock Policy Officer at the FAO, who co-led GLW3, answers a few questions about this new data.

Q: Why map livestock? 

A: These data are crucial to underpin the analyses, monitoring and evaluation that will be needed in support of sustainable livestock sector development.

Livestock bring many benefits to humanity, including providing us with nutritious food and micronutrients, enhancing livelihoods, and building resilience against climate and economic shocks. But there are also risks associated with a burgeoning global livestock sector. Livestock feed production competes for land with crops grown for direct human consumption. Intensification raises animal welfare concerns and can increase the risks of zoonotic and food borne diseases. And there is an urgent need to address increasing pressure that livestock production exerts on the environment through land use change, feed production, overgrazing, biodiversity loss and greenhouse gas emissions.

Production of animal-source foods is only increasing in response to growing demand, particularly in low-and middle-income countries. If we are to continue reaping the many benefits that livestock offer humanity, whilst mitigating these important risks, we need the right institutions in place, policies that are based firmly on analysis and evidence, and information systems that can be used to measure baselines and monitor progress towards the 2030 Agenda for sustainable development.

Q: What are the main challenges with collecting this type of data?

A: Originally we faced a technical challenge of disaggregating the data, finding the models that work, accessing enough computing power. Over the three versions of GLW, we’ve pretty well nailed these problems. Now the limiting the factor and biggest challenge is the quality of the input subnational data. We don’t have the resources to go into every country and work with national partners, so there is a lot of following up by email and phone. We collect census data and national statistics from a variety of sources such as national veterinary departments, statistical departments and yearbooks. In large countries such as India and China, data tend to be collected at state level, but not centrally, which further complicates things.

Q: What are some of the most important improvements you’ve made for this update?

A: The latest version of GLW improves on the 2007 and 2014 iterations in a number of ways.

• National data have been greatly updated. There are important improvements in the census data on which the models are based. A good example is for China, which makes a big difference to the overall numbers. Countries are collecting better data and are becoming more open to sharing.
• We adopted a different modeling approach. We used to use stratified regression for modeling, and we’ve replaced that now with random forest models. This is a machine learning approach that is very flexible, less constrained by correlations across predictor variables versus the regression models, and which allows multiple interactions between predictors. Random forest makes predictions using multiple decision trees, and we’ve demonstrated that it provided much better predictions than the previous modelling approach.
• Another major improvement is that we’ve made our data catalogue fully open-access for the first time using dataverse. And we’ve added more detailed and better-quality metadata that offers details on the quality of the models. Researchers will be able to see the uncertainties around input data, the average size and age of the census units, and see where in the world we have very old or new data. As well, the new livestock systems website has been set up to help communicate the results to different types of users.

Q: What are some immediate applications of the data?

The data can be useful to anyone from a high-level politician to a subnational program implementer.

A high-level person could use the maps in presentations in order to convey important messages and facts about the number and location of livestock around the world. We’ve also developed some Key Facts graphics that can be used for this purpose.

At the subnational level, the data could be deployed to inform many kinds of programs for livestock health and productivity. They can be linked to epidemiological studies for example, as reliable information on the distribution and abundance of animals is essential for responding to diseases such as Avian influenza.

As well, international agencies such as the World Bank and FAO can use these data to spot gaps in agricultural information, and better direct their funding for agricultural censuses. 

Public-private partnerships such as GALVMed, which develops vaccines for major livestock diseases impacting smallholders, could make use of these of maps to determine the quantity of vaccines that need to be distributed.

Q: FAO is part of the Livestock Data for Decisions (LD4D) Community of Practice. How can the community benefit from this tool? And how could the community contribute to its ongoing development?

A: I encourage the community to use and cite the data and publish their results in high-impact journals. FAO and ULB are keen to be involved as a partner in projects that use the data, as there are many unpublished developments that go beyond the open-access data. As well, if another organization has access to national data, we urge them to share these data sets with us. Finally, we want expert feedback on the quality of inputs to the models, on the predicted distributions, and on their use.

Q: What is the next big thing, in terms of data and modeling, that will help improve these data for the next update?

 A: Our big ambition at the moment is to convert the animal numbers into the amount of meat, milk and eggs produced – now, and in the future, under different scenarios of growth. This goes hand in hand with improving our maps of production systems and we are now equipped to do this well.

We are also working on time series. Because the different versions of GLW have different reference years (2005 and 201), people naturally want to use them as a time series. But we warn against that because methodologies have changed and they are not directly comparable. However, we could try to apply the current approach retrospectively, and look at past trends, as well as thinking about how things might look in the future.

There is a particularly close link between the GLW data and GLEAM – the Global Livestock Environmental Assessment Model. GLEAM is a tool developed by FAO to estimate the environmental impacts of the global livestock sector, such as greenhouse gas emissions and natural resource use efficiency. Better livestock statistics underpinning GLEAM will give us more robust estimates of these environmental impacts. All of these applications will be made available on the Livestock Systems website.

Key links

Gilbert. M. et al. 2018 Global Distribution Data for Cattle, Buffaloes, Horses, Sheep, Goats, Pigs, Chickens and Ducks in 2010. Sci. Data. 5:180227. doi: 10.1038/sdata.2018.227.

UN Food And Agriculture Organisation (FAO) Livestock Systems

Written by Vanessa Meadu, Communications and Knowledge Exchange Specialist for the Supporting Evidence Based Interventions programme based at the University of Edinburgh. This is an edited and condensed version of an interview with Tim Robinson, UN Food and Agriculture Organisation (FAO). Additional input from Marius Gilbert, Université Libre de Bruxelles (ULB). SEBI and FAO are members of the Livestock Data for Decisions (LD4D) Community of Practice.

Feature image credit: ILRI (source)