Open access publications in the SLU publication database

Abstract

The Swedish official cultivar testing conducts multienvironmental trials (MET) to makerecommendations of cultivars that are well adapted to farmers’ regional conditions. Inthe MET, a large number of cultivars are tested in several geographical regions. Thetested cultivars perform differently in varying soil types and climates, a phenomenonknown as genotype×environment interactions. The MET data structure is often large andhighly imbalanced, which causes computational problems when applying some statisticalmethods. Several issues, such as prediction of crop variety performance and efficientcomputation of measure of cultivar stability are urgent to be tackled by developingcomprehensive and robust statistical methods. This study aims to address these issuesand provide a gold standard for MET analysis in Swedish official cultivar testing.

In this study, we investigated several linear mixed models by using cross-validation(CV). We proposed to use random cultivar effects, known as best linear unbiasedprediction (BLUP) method to replace the current fixed cultivar effects, known as bestlinear unbiased estimation (BLUE). In theory, BLUP provides more accurate rankingsand predictions than BLUE. The current-practice analysis strategy, i.e., two-stageunweighted strategy, was also compared to several strategies such as single-stagestrategy and two-stage weighted strategies that comprise some weighting methods. In theCV, mean squared error of differences (MSEP) was used to assess the performance ofestimation of cultivar effects by BLUP and BLUE to select a model that provides bestprediction accuracy. A new inter-zone stability measure was also proposed to tacklecomputational burden and provide additional useful information regarding cultivarstability across zones and years.

The MSEP revealed that BLUP outperformed the current-practice method, BLUE,and so improved the accuracy of zone-based prediction. Also, the single-stage and twostage weighted strategies outperformed the current strategy. The proposed stabilitymeasure offered a less computational resource, and provided more flexible stabilitymeasure for practical purpose.