Traditionally, agricultural experiments are conducted on small plots or in greenhouses. However, from the farmers' perspective, they want to know what is happing on a particular paddock. The challenge in large on-farm experiments (OFE) is to accommodate spatial variation. We are running a few statistical analyses about the topics of OFE

Principal investigator

Zhanglong Cao zhanglong.cao@curtin.edu.au
Magnifying glass

Area of science

Agricultural And Veterinary Sciences, Statistics

Systems used


Applications used

Partner Institution: Curtin University| Project Code: A000334

The Challenge

Computing Bayesian inferences and genotype imputation is extremely time-consuming, which is the challenge of implementing these techniques on agricultural data sets.

The naïve posterior sampling method took us two weeks on running 8000 interactions/sampling of correlated parameters from multi-dimensional space.

The Solution

To overcome the challenge, we optimised and vectorised our sampling methods, and used parallel sampling approaches to replace simple sampling approaches. We also deployed the heavy computing tasks onto the Nimbus instance, and export the outcomes into our laptops for post-computing analysis and visualization.


The Outcome

The computing time was reduced from two weeks to 30 minutes on running 2000 interactions by 4 chains. In the meantime, with the help of the Nimbus instance, we save resources on our laptops and are able to concentrate on statistical modelling and manuscript writing, rather than worrying about system crashes caused by low memory or disk storage.

We also became more ambitious to run genotype imputation, factor analytic on large data set and grid-search techniques on REML models.

List of Publications

Bayesian Inference of Spatially Correlated Random Parameters for On-farm Experiment, Field Crop Research.

Optimal design for on-farm strip trials — systematic or randomised? Ready for submission.