Abstract

The description of the production cycle of crops with multiple harvests, such as strawberries, can be conducted using non-linear regression models. When there is a lack of homogeneity of variances and non-normality of errors, an alternative is to use non-linear quantile regression. Thus, in this study, the objective was to propose more robust estimates of the parameters of the growth models via a non-linear quantile regression model, with greater precision and parsimoniousness, using the Logistic, Gompertz, von Bertalanffy, and Brody models with the variables cumulative number and cumulative mass of strawberry fruits. The work was conducted using data from two experiments conducted in Santa Maria, RS, and Brazil. In Experiment I, the treatments were four strawberry cultivars: LBR F, LBR, Albion, and Estiva; in Experiment II, there were three strawberry cultivars: Albion, Estiva, and LBR. The non-linear regression models Logistic, Gompertz, and von Bertalanffy for the variables cumulative number of strawberry fruits and cumulative mass of strawberry fruits, both by the ordinary least squares method and by the non-linear quantile regression method, were those that presented the best results in the quality of fits. The Logistic model showed high accuracy in Experiment I (in the central quantiles) and maintained a dominant performance in Experiment II (between 0.44 and 0.70), even with greater data variability. The use of non-linear quantile regression is recommended as an alternative for fitting the Logistic, Gompertz, and von Bertalanffy models in data with a sigmoidal distribution when the assumptions are not met. The NLQR showed a drastic reduction in residual deviations compared to the OLS, being between 35 and 2,450 times more accurate in the experiments.

Keywords

assumptions violation, multi-harvest crops, quantile modeling