This project will use earth observation combined with local ground data to develop machine learning algorithms to estimate livestock numbers in East Africa.
Livestock are a key component in food security and livelihoods in pastoral communities across Africa. Feeding livestock is increasingly challenging for pastoral communities due to the dual pressures of climate change and human population increases. Furthermore, estimates of greenhouse gas emissions from the livestock sector are hampered by inadequate information on livestock numbers and feed consumption, both important elements of prediction algorithms. This project will feed into the wider work of the Jameel Observatory for Food Security Early Action by helping predict impending shortages of livestock feed which lead into human food crises. Estimating livestock feed availability in relation to livestock population is a crucial activity for the Observatory if we are to better respond to the impacts of climate change on food security. Currently neither the feed availability nor the livestock population are easy to estimate. Work is ongoing within the Jameel Observatory to estimate livestock feed availability and this project will work on estimating the livestock numbers to estimate feed demand. Existing approaches such as the Gridded Livestock of the World layer mainly use agricultural census data which can be unreliable. This project will examine how new data sets such as Earth Observation satellite imagery along with AI and machine learning methods could be used to estimate livestock numbers more accurately at more frequent time intervals.
In this project we will combine expertise on livestock estimation, modelling, machine learning and satellite image analysis to support a student to develop new ways of estimating livestock numbers in East Africa. The project will build on existing work to estimate livestock numbers using, very-high spatial resolution satellite data (<5m) available through the wider project, UAV data, agricultural census data and household survey data as well as publicly available Earth Observation data. The UAV data will be collected as part of the PhD with support from the Jameel Observatory. The student will explore the use of machine learning algorithms for classifying very-high spatial resolution satellite imagery and ground-truthing spatial data through a combination of primary household surveys and interrogation of existing secondary data on livestock in East Africa. Estimates will involve developing predictions of livestock using AI and machine learning algorithms.
The FAO has developed the Global Livestock Environmental Assessment Model (GLEAM) which uses crop distribution and yield data in combination with data on types of feed. However, these estimates currently rely heavily on expert opinion and the model is currently unable to account for seasonality in feed availability. Improvement of the accuracy of feed basket estimation and accounting for seasonality would significantly enhance the ability of the model to provide realistic estimates of GHG emissions from the livestock sector which could feed into global climate modelling. The improved estimates from this PhD will be incorporated into GLEAM on an experimental basis with the possibility to use the newly developed approaches for a wholesale revision of the feed basket component of GLEAM.
Key Research Questions
- Can livestock numbers be accurately counted using a range of metrics derived from Satellite imagery?
- Can UAV data and machine learning approaches generate estimations using freely available Sentinel-2 10 m resolution data?
- Can these predictions be used to develop spatial layers of livestock composition?
The project will be split into a series of components as follows:
- During the first year the student will assemble secondary data on livestock in East Africa using datasets available from within the supervisory team and their extended network of partners and collaborators and publicly available data. Multi-resolution Earth Observation data will also be assembled. A literature review will be used to identify appropriate datasets and methods for the project and for the student to adapt the research questions to these literature findings and their own interests.
- During the second year the student will construct predictions of livestock based on a series of hypotheses about what influences livestock numbers and locations which will be developed with the expert supervision team and the literature review outcomes. These predictions will subsequently be applied to generate estimations of livestock population for selected locations in East Africa.
- During the latter stages of year 2 and into year 3 the predictions will be ground truthed by using primary household survey data. Year three will culminate in experiments with incorporating livestock estimates with livestock feed-basket maps in order to improve predictions emerging from the GLEAM model.
The student will join an interdisciplinary research group combining researchers in the Vet School, the Global Academy of Agriculture and Food Security and the school of Geosciences working on socioecological systems in Africa which will provide extensive peer support. This will be complemented by specific training from the supervisors in remote sensing, statistical analysis and fieldwork skills. The FAO will provide training in Gridded Livestock of the World layer. The Jameel Observatory is a Community of Practice including a range of food security partners in East Africa. There are also formal training events in all these topics that may be accessed as needed. A comprehensive training programme will be provided comprising both specialist scientific training and generic transferable and professional skills
The project would suit a student with a degree in Informatics, computer science, natural science. Support will be given to a student with an interest in ecosystems, remote sensing and natural resource use with an undergraduate or masters in any quantitative subject. A background in data science, multivariate statistics and spatial analysis is advantageous. A willingness to learn new methods and work effectively in LMIC environments, and work with people from different backgrounds.