Comparison of Spatial Models for Sugar Beet and Barley Trials

Abstract

Several spatial methods exist for the adjustment of local trend in one dimension. The aim of this study was to evaluate and compare the precision of different spatial methods. For this purpose, 293 sugar beet (Beta vulgaris) and 64 multienvironment barley (Hordeum vulgare) trials of two German plant breeding companies were analyzed using a baseline model, which comprised a block and replicate effect, and different one‐dimensional spatial models augmenting the baseline model. Model fit was assessed using the Akaike Information Criterion (AIC), the phenotypic correlation of the adjusted genotype means between two environments, and the relative efficiency. For the sugar beet and barley trials the baseline model outperformed the spatial models in the majority of cases, while in some cases the addition of a spatial component proved beneficial. Based on these results we propose a conservative approach to spatial modeling that starts with a baseline model and then checks whether adding a spatial component improves the fit. Among the alternative models studied, the linear variance and the first‐order autoregressive models were the most promising candidates.

Can Spatial Modeling Substitute for Experimental Design in Agricultural Experiments?

“…This design has been shown Table 2. Experimental designs that exerted some local control on the spatial variability, like ALPHA or row-column designs (Müller et al, 2010;Sripathi et al, 2017), generally increase the power of the ANOVA (Legendre et al, 2004). This as superior to RCBD (Masood et al, 2008;Gonçalves et al, 2010) due to stronger local control.…”

“…Müller et al (2010) found that in barley (Hordeum vulgare L.) and sugar beet (Beta vulgaris L.) trials, the baseline model, which only includes a block and a replicate effect, showed the best fit according to Akaike information criterion for most cases. Müller et al (2010) found that in barley (Hordeum vulgare L.) and sugar beet (Beta vulgaris L.) trials, the baseline model, which only includes a block and a replicate effect, showed the best fit according to Akaike information criterion for most cases.…”

Spatial Variability Effects on Precision and Power of Forage Yield Estimation

“…Among two-dimensional AR1 ´ AR1 n models, the mean autocorrelation estimated among rows was smaller than the autocorrelation among columns for trial ´ year (0.61 vs. 0.96) and trial (0.66 vs. 0.86) analyses. Müller et al (2010) mentioned that two-dimensional models were not effective for trials designed with two and four columns. In analysis of spatial variation in Commonwealth Scientific and Industrial Research Organisation (CSIRO) cotton (Gossypium hirsutum L.) trials, Liu et al (2015) observed a higher frequency of spatial models in rectangular trials compared with square trials.…”

A comprehensive comparison between ANOVA and BLUP to valuate location-specific genotype effects for rape cultivar trials with random locations

“…Clearly, the presence of genotype-location interaction effects will attenuate the correlation relative to a situation with no interaction. Since the operating interaction effects are the same for both analysis methods, however, the correlations can serve as a relative measure of precision for different analysis methods (Müller et al, 2010).…”

“…The efficiency of analysis methods was determined as described by Wu et al (1998), Wu and Dutilleul (1999), Qiao et al (2000) and Müller et al (2010). According to the method in these literatures the average variance of differences (v.d.…”