Consideration Of Field Heterogeneities In The Calculation Of Variety Means From Agronomic Trial Data: Comparisons Between A Nearest Neighbor Adjustment Method And Experimental Designs Like Lattices And Latinized Alpha-Designs

Agronomic experimentation aims at testing and comparing a wide number of varieties in different environmental conditions in order to select the ones best adapted for a given purpose. A major goal is to establish a ranking of the varieties, based on several variables measured on the plots and on some possible field heterogeneities that may bias the raw results.

The knowledge of the experimental sites generally guides the choice of the experimental design to use: designs without varieties’ means adjusting (e.g. randomized complete blocks) if the field can be considered as homogeneous, designs taking gradients into account (e.g. lattice or alpha designs) otherwise. Yet, the latter designs impose strong constraints on the shape of the trials and on the possible numbers of varieties and replications.

Labkey software currently offers a set of experimental designs covering these common adjustment needs for the removal of unidirectional or bidirectional gradient effects. In order to free ourselves of their constraints, we explore here the possible advantages of using Schwarzbach’s algorithm, a nearest neighbor adjustment method. This poster presents comparisons between this method and available Labkey designs, on artificial datasets of twenty-five varieties in four different environments: homogeneous field, unidirectional gradient, bidirectional gradient and non-structured gradient. Two criteria were used to draw conclusions: the ability of the algorithm to recover the true ranking of the varieties and the closeness between their estimated and actual means.