Model Comparison for Multienvironment Trials

In multienvironment trials, the local spatial tendency within trials and the residual heterogeneity between trials can be jointly modeled in the context of mixed linear models. The objective of this study was to know the suitability of spatial models within trials and the residual heterogeneity between trials for multienvironment fertilizer trials. The research used field trials data of rice fertilization in Karawang and Kebumen region. Fertilizer treatments consist of fourteen levels of fertilizer type with three replications. In Kebumen region the research was conducted over two growing seasons. The experiment used a randomized complete block design. The results show that for single experiment the best model for Karawang and Kebumen1 location was RCBCE power isotropic model, whilst for Kebumen2 location was RCBCE power anisotropic model. In combined experiments the best model was CE power anisotropic model with heterogeneous residual variance across locations. Modeling of the local spatial tendencies using an analysis of variance including heterogeneous residual variances increased the ability to identify differences among treatments.