Spatial Mixed Model Analysis in Varietal Selection Field Trials

Abstract

Spatial variation is common in varietal selection field trials and is a central problem confronting a plant breeder when comparing the varieties' genetic potential. If spatial variability is not taken into account, it can strongly bias variety estimates and result in large standard errors. There have been many methods developed for accounting for spatial variation. Of these, the spatial mixed model approach proposed by Gilmour et al. (1997) has received particular attention as it simultaneously considers three types of spatial variation to be modeled: local, global, and extraneous variations. Despite the recommendations by several authors, spatial mixed model techniques are not widely used in the crop variety evaluation program as a routine data analysis platform. We  present a spatial mixed model analysis using field trials from Ethiopia. Results of spatial analysis are compared  to that of randomized complete block (RCB) analysis. The investigated spatial mixed models show better data  fitting, resulting in a smaller error variance than that of RCB model analysis and a substantial improvement in  heritability. Thus, spatial mixed models must be routinely employed in analyzing field trials to accurately and  efficiently select superior varieties that contribute to agricultural productivity.