Skip to content
menu icon

Students unlock new statistical models to help analyse field data

Argentinean students Lucas Peitton and Eugenia Settecase were in Australia last year to work with the Statistics for the Australian Grain Industry's northern node.
Photo: Toni Somes

A new project is helping to build a statistics workforce for the Australian grains industry.

A new statistical model could soon be more widely and confidently used in variety trials to account for in-field spatial variability - with the added potential for quicker and more cost-effective results.

This is because of research by Argentinean student Lucas Peitton, who was in Australia in 2019 to work with the Statistics for the Australian Grains Industry (SAGI) northern node.

Mr Peitton's time in Australia, to explore new statistical modelling, was part of a broader initiative, says GRDC enabling technologies officer, John Rivers.

Attracting and building a workforce in statistics for the Australian grains industry has been an ongoing challenge.

"So, a joint project between GRDC and the Queensland Department of Agriculture and Fisheries (DAF) set out to develop international partnerships and build capacity in the growing field of agricultural statistics," Mr Rivers says.

This project sets a precedent for student work placements and graduates applying for jobs within SAGI into the future. - GRDC enabling technologies officer John Rivers

That project supported two honours students from the Statistics Department of the National University of Rosario, Argentina - Lucas Peitton and Eugenia Settecase.

Both worked with the SAGI team at DAF last year - completing valuable research projects under the supervision of senior biometricians Gabriela Borgognone and Valeria Paccapelo.

In-field spatial variability

Mr Peitton set out to better understand a new statistical model and its implications for managing spatial variability in yield trial data.

Spatial variability - the variability observed in yield due to changes in fertility, moisture or management practices - can be present in field trials.

It needs to be accounted for using statistical models so that reliable predictions of variety performance can be made.

This is traditionally done using linear mixed models and following a multi-step process, which includes graphical diagnostics and formal tests of hypothesis.

It also requires a basic understanding of these models, which can preclude some in the industry from using them.

On top of that, these can be time-consuming for routinely analysing trials - particularly when they are many.

Newer to the world of statistical modelling is SpATS - or Spatial Analysis of field Trials with Splines model - which was created in 2016 and shown to perform as well as the traditional method when analysing large sorghum trials.

Mr Peitton was tasked with determining whether the model could be used for smaller trials and other crops.

He analysed more than 30 trials of varying sizes from the GRDC National Mungbean Improvement Program and National Chickpea Breeding Program and found the SpATS model worked well. It showed consistent results compared to the traditional model.

Although it too uses a linear mixed model, people who wish to use it do not need to be experts.

Its single-step approach also makes it very fast for routine spatial analyses - a general model can be obtained within seconds.

It also has additional benefits. It uses standard code and its statistical computing and graphic package is free. This compares to the traditional model, which requires an annual licence to be paid.

Mr Peitton, who is now back in Argentina, says the model's performance paves the way for it to be more broadly used.

"Now that we know that the SpATS model performs well, with appropriate guidelines it could be used more widely by agronomists and other researchers who run varietal trials," he says.

"It could help them to obtain better varietal predictions than using simpler methods that do not account for spatial variability in the field."

Multiple trait analysis

While Mr Peitton was testing this newer statistical model, fellow student Eugenia Settecase was working to better understand multi-trait models.

In standard variety trials - where the aim is to identify the best-performing genotypes - several traits of interest, such as yield, disease resistance or quality characteristics, are analysed separately.

But, measurements for more than one trait are often taken on the same experimental unit - be that a field plot or pot in the glasshouse.

Although multiple traits can be studied separately using several approaches, none of them account for the correlation between measurements taken on the same experimental unit.

Yet, analysing these correlations can help to distinguish between a trait's genetic contributions and external factors.

Ms Settecase's work focused on extending a statistical model that was originally developed for analysing two traits simultaneously.

This extension provides a model suitable for datasets with a small number of traits, estimating each trait's genotypic effects and accounting for the correlation between them. If the experiment has spatial variability, that can be modelled, too.

Her work used data from the National Mungbean Improvement Program's field experiment carried out at the Hermitage Research Facility, Warwick, in 2018.

Ms Settecase evaluated the diversity of a set of genotypes in terms of two different groups of characteristics - one related to yield and another related to seed quality traits.

These models led to more accurate estimation of genetic effects, particularly for traits whose measurements are heavily affected by field trial 'noise'.

This will help the NMIP make better selections of lines for further breeding.

International links

In December 2019, Mr Peitton and Ms Settecase were awarded for their work.

They presented posters with the results of their research topics at the International Biometric Society's Australasian Region Conference in Adelaide. Mr Peitton won first prize and Ms Settecase was runner-up.

Mr Rivers says that not only was the students' work successful, but so too was the broader initiative to attract and build a statistics workforce.

"This project sets a precedent for student work placements and graduates applying for jobs within SAGI into the future. It has also provided important links with other international researchers."

GRDC Research Code DAQ1905-003

More information: John Rivers,;

back to top