Large data to underpin national livestock food systems modelling

Challenges

The livestock industries are continually under pressure to proactively embrace and incorporate sustainability drivers into their systems and forward improvement plans. Previous work on genetic improvement has shown a very high return on investment (£2.4 billion over 20 years). There are several barriers to genetic improvement, one of which is the availability of sufficient phenotypes of any given quality. This is even more evident in a genomic selection framework whereby genotypes are no longer limiting and greater exploitation of existing genotypes could be obtained with more and novel phenotypes.

We have previously been able to demonstrate the value of incorporating new data sources into these programmes. We developed novel methods of cleansing and merging structurally disparate and unstructured datasets which required the development of new algorithms and data systems (rules and logic) to extract new knowledge not previously possible. These systems were then harnessed to analyse new and existing cattle traits in a much larger, and thus better-powered, dataset, and thus generate more reliable predictions of offspring traits to aid the selection of animals for breeding.

This innovation represented a step change as before this, genetic relationships between individual animals were only available for animals registered with a pedigree breed society, which accounts for less than 40% of UK dairy cows and 10% of UK beef cattle. In addition, beef genetic evaluations were previously based on performance data from registered pedigree herds. This data represents less than 5% of the total beef population. and less representative of the goal of improving production traits in commercial herds and of lower accuracy.

Integrating data from various sources has meant that information from 80% (5 million) more animals are now available for research and livestock improvement tools, enabling breeders to incorporate new traits of interest and analysis techniques into genetic improvement programmes. By working with these extended datasets, we have been able to research a vast array of traits that underpin efficiency and sustainability in dairy and beef systems. This research has enhanced our understanding of:

• Calf survival: early life survival in beef and dairy cattle is heritable
• Carcass traits: By merging British Cattle Movement Survey and abattoir data, we have shown that carcass weight, finishing age, carcass fat and protein are all heritable traits in beef cattle

Further work is required to build on these and other examples of livestock-related “big data” to explore how large-scale modelling of these data can help us understand how our complex livestock systems and sectors behave and respond. 

Questions

Solutions

This project aims to demonstrate how robust and secure collation of data from a range of public and private sources can drive big data research in the Open Science era and also create the space where the industry can engage and identify impact and exploitation opportunities. To do this we are exploring pilot examples of how animal-centric data from disparate sources can be gathered into a single cohesive dataset used to drive large-scale modelling and generate recommendations to inform policy.

Role of remote images in beef supply-chain efficiency

There is a great degree of variability in the beef supply chain. This is largely due to a lack of consistent and accurate data collection and subsequent analysis leading to unacceptably high disease levels, poor business performance, and huge variability in product quality. We are understanding the drivers of agricultural productivity, product quality and environmental sustainability within the beef supply chain. We are developing data-driven solutions to address potentially conflicting challenges for beef producers and the whole supply chain.

We are also collecting hyperspectral images of the loin on up to half of the animals to explore predictions of key meat quality traits. These images have been demonstrated to measure gross components of the product but also quantification of non-visible components, such as chemical composition or muscle structure. We will pilot if machine learning approaches of these images could be used to provide a remote assessment of animal and system efficiency in a beef supply chain.

Reusing animal-centric data to derive industry performance indicators for industry and national statistics and reporting 

Building on the reuse of data to provide animal-centric metrics for UK Greenhouse Gas Inventory reporting, we are exploring these data for further enhancements to the UK inventory, including developments in smart inventory approaches including integration of spatial and weather data. We are refining our metrics of industry-level animal performance that help to fuel the UK Greenhouse Gas Inventory to explore temporal and spatial variation and the role of weather on these indicators. We are also modelling the data to explore if we can estimate likely representative farm system type allowing us to explore animal and system variation and interactions. If successful this could lead to new ways of routinely reporting across a range of topics, measures of key performance indicators at industry and national levels, and feed into wider modelling of technical, economic and environmental efficiency potential of the Scottish livestock landscape.