Spatial analysis on potato breeding trials

Abstract

Field experimentation for plant phenotyping is one of the main plant breeding activities. Estimating the genetic parameters of interest are determining factors for the success of the breeding program. Some errors can happen in this process: differences in soil fertility, the quantity of light intercepted, plot competition, errors on phenotyping, abiotic and biotic stress not homogeneous on the area, all these effects can change the process of selection. The use of good analytics approaches is a good way to reduce those errors. This study checked the efficiency of spatial models, using two different approaches: first-order autoregressive (AR1) and SpATS model. We used 30 potato breeding trials from the Universidade Federal de Lavras potato program (PRO-BATATA) and the University of Florida. The trials were carried out between the seasons 2016 - 2019, with the complete random block and augmented blocks designs. Three traits were used: total tuber yield (TTY), marketable tuber yield (MTY) and specific gravity (SG). On the first stage, the data were analyzed using a mixed models approach with independent errors.After this the data were analyzed using the models AR1xAR1, AR1xAR1+nugget and SpATS. To compare the results, we used the prediction error variance (PEV), heritability (h²) and also the impacts on the selection of the best genotypes. The models based on the AR1process got better results when compared with the base model in 81% cases for TTY and for MTY, and 61% for SG. There is not an unique model for all trials. It was necessary different parameters to control global errors in each case. Both models, AR1 and SpATS, were successful to control the local and global errors, showing the same results. The relative efficiency using the AR1 model were 119, 113 and 107% and for the SpATS 121, 116 and 118%, for TTY, MTY and SG respectively. Spatial models showed worst relative efficiency in just few cases. The average heritability was higher using the spatial model for all traits.These differences were 14% for TTY, 8% for MTY and 7% for SG. There were differences in the selection of the best clones using the spatial models in cases where the spatial variation was medium to high.These results demonstrate the importance of using the most accurate analytical method. Spatial models were more efficient to control the field variation compared with the model that assumes independent errors. AR1 and SpATS do not showed differences in their results. We recommend using the spatial analysis, based on AR1 or SpATS, as a default to analyze field data from a potato breeding program.