Spatio-temporal modelling of environmental and crop management effects on sensory scores and chemistry profiles using hi

Host Faculty: Engineering
General Subject Area: Statistics and Data Science
Project Level: PhD
View the Research Website
FIND A PROJECT - Engineering

Plant and Food Research Ltd has access to large historical datasets of consumer sensory scores and chemistry profiles on multiple individuals for a combination of management-environment interactions for various commercial varieties of crops such as hops, apples and kiwifruit. Utilising the available phenotypic data at Plant & Food, we will develop a modelling strategy to predict environment x management x consumer interaction effects. We focus on selecting sustainable varieties, adapted to climate change and a low environmental impact. Consumer data allows us to explore economically viable options for smaller and more specialised market segments. Predicting consumer specific effects can include the analysis of high dimensional chemical profiles or discrete panel data.
Including ExMxC interactions in a plant breeding model comes with several challenges. Predicting all combinations of different growth conditions and market segments will lead to a large set of model parameters, which requires a large amount of information, usually obtained from combining the data of several breeding trials. This can result in the need to combine outcomes for different sampling designs and missing data where not all full-factorial combinations are available.
Using a Bayesian framework will allow us to use historical information and expert knowledge about phenological effects. A Bayesian model combines the information from the data with a prior distribution that quantifies the belief about a model parameter before any data is collected. This mechanism allows us to evaluate the full posterior distribution of the effects of interest. We will use numerical procedures, like Markov Chain Monte-Carlo algorithms, to sample from these posterior distributions using the statistical software R and Stan.
Most models need to include spatial effects for adequately representing the environmental variability, modelling the spatial correlation between different sampling locations. Our aim is to look at different locations that are analogous to future climate developments to predict if a genotype is suitable under global warming or extreme weather events.


Supervisor: Daniel Gerhard

Key qualifications and skills

Good programming skills and Bayesian statistics

Does the project come with funding


Final date for receiving applications



Applied statistics