Spatial modeling increases accuracy of selection for Phytophthora infestans-resistant tomato genotypes

At initial breeding stages, using a replicated check design is a viable alternative to reduce experimental field area as well as financial and operational costs. In this situation, spatial modeling could act to increase prediction accuracy of plant genotypic values. The objectives of this study were to demonstrate how spatially adjusted models can be used to reduce experimental error and how to compare statistical models in order to identify the best model for accurate genotype selection. For this purpose, we assessed 200 F-3:4 tomato families for their resistance to Phytophthora infestans isolates. NC1CELBR, NC25P, and the cultivar Santa Clara were used as checks. Under field conditions, plants were inoculated with P. infestans isolates and scored according to their level of disease severity. Nine statistical models were adjusted to estimate family genotypic values. The selection of the fittest model was based on residual variance values, accuracy, Akaike and Bayesian information criteria, and the maximum likelihood ratio test. We observed spatial patterns within the experimental field area. Spatial modeling decreased error, which is indicated by the better experimental variation distribution. Residual variance decreased, while genotypic variance increased similar to 10% when spatial analysis was used. Spatial analysis improved selection accuracy by 19% compared with the traditional analysis. Therefore, we recommend incorporating spatial modeling into data analysis in breeding trials for disease resistance because it can provide higher gains from selection compared with traditional modeling approaches, depending on the experimental condition.