Predição de produtividade para seleção de híbridos de tomate de mesa

Abstract

The fresh table tomato stands out in Brazil as a vegetable of great economic importance for the value of production and also socially for the generation of jobs. In this context, the companies seek the development of new hybrids, which among other qualities, are productiv e in the different regions of fruit production. This research and development work generates a high cost of travel and food for the companies, due to the need of its employees to be present in the different places for harvesting and evaluation of hybrids p roduction. The objective of this work was to propose a model to evaluate the yield of fresh tomato hybrids that is made with better use of financial resources. For that, three experiments were carried out in the years 2016 and 2017 at the Experimental Station of Syngenta Crop Protection in Holambra - SP. Model fit was made based on the results of experiment 1 (training data), in which the best models were selected and tested with data from experiments 2 and 3 (test data). For each of the response variables studied (PCA, PCAA, PCAAA, PC23A and PMF), multiple linear regression models were fitted for all possible combinations, from 1 to 11, of the independent variables, collected. The selection of the best models was based on the graphical method, in which the gain in the coefficient of determination (R2) was weighted as a function of the increase in the number of tomato cluster that were added to the model. The selected models were then evaluated in the test database using the parameters: correlation, mean abso lute error and mean absolute percentage error. In order to verify the possible occurrence of significant differences among the genotypes, the means of the variables commercial fruit weight AAA (PCAAA), AA (PCAA) and A (PCA), AA + AAA (PC23A) and average co mmercial fruit weight were submitted to analysis of variance and later, compared by the Scott -Knott clustering test (5%). After this analysis, the clustering of the genotypes with the predicted data was compared with the clustering of the genotypes with the real data to verify the accuracy of the estimated data. Among the variables studied, PCAA and PC23A presented reliable estimates, considering the low percentage absolute errors between real and predicted data, around 5%, and also because they presented similarity in the clustering by the Scott -Knott test. The PMF variable, although it did not present the same order in the cluster by the Scott Knott test in both experiments, is also feasible to be used as a decision factor in the selection of tomato hybrids, since the mean percentage errors were 3.49 and 3.62% in experiments 2 and 3, respectively. It was concluded that the model with four clusters (2, 3 6 e 9) is the best to estimate fresh tomato hybrids yield.