Using Correlated Data to Improve the Analysis of High Throughput Phenotyping Trials.

The evaluation and characterization of plant genetic resources frequently occurs at trial sizes much larger than standard physiological or breeding evaluation trials. The experimental designs for such trials need to consider the logistics of handling large entry numbers whilst maintaining statistical design integrity. Furthermore, the data collation processes are necessarily different from standard trials and hence different approaches to analysis need to be considered. In this study we present the experimental design and analytical techniques implemented to evaluate trials with more than 10,000 entries. The experimental designs are modified augmented grid-check designs where the aim is to enable modelling of the spatial variation across these ~1ha trials. These large trials necessitate high throughput phenotyping techniques to capture data in an efficient manner. The subsequent data structures are split or correlated and require different statistical approaches. Two approaches are presented to analyze data measured at irregular intervals derived from a single procedure such as remote sensing imagery. Both, time series analysis and multi-trait mixed models to accommodate between trait correlations were shown.