Amostragem sequencial bayesiana para estimação do vigor e danos das sementes de soja através do teste de tetrazólio

Abstract

Soy is an oilseed of great relevance to Brazil because it is one of the main crops produced in the country. The success of this crop depends on several factors, the main one being seed quality. For quality control, the tetrazolium test has stood out because in addition to assessing the germination viability and the vigor of the seed lots, it makes it possible to classify the types of damage. Sampling is an important step that should be reduced in cost. Thus, the use of sequential sampling, which takes into account the immediate evaluation of each seed, can allow the decision on viability and vigor to be taken in advance, without all the seeds being evaluated as foreseen in the classic test due to the formality of the method. In addition, it can improve the estimates obtained through Bayesian techniques, including a priori information. Thus, the objective was to study the application of Bayesian sequential tests to verify the vigor and damage proportions by categories in soybean seeds through the tetrazolium test. Binomial modeling was used to estimate the proportion of vigor in batches and Multinomial for damage categories (mechanical, moisture or bed bug). In each case, the combined priorities were used and the parameters were elicited. It can be concluded that the four frequentist, Bayesian, sequential and sequential Bayesian approaches, were efficient to perform the estimation and decision making on the parameters, with a reduction in the sample size. In the sequential Bayesian process, both in the estimation of vigor (use of the binomial distribution) and of damage (use of the multinomial distribution), it was possible to reduce the sample size. Moisture damage was present in 20.16% of the evaluated soybean seeds, by bed bug in 3.5% and mechanical damage, in the case of manual harvesting, in only 1.92%. It was also obtained 1% of the seeds with more than two damages.