Regressão linear simples e múltipla para predição de colheita em bananeiras Tipo Prata

Abstract

One of the demands of Agricultural Science is to generate simulation models for growth and development. In banana crop, this knowledge makes possible to estimate the production of a certain plant from other attributes, which should preferably be of practical application, easy determination, with use of simple tools, and prior to harvest, which subsidizes an effective harvest program. This study aimed to adjust regression models for predicting harvest of banana type 'Prata'. The experiment consisted of trees and bunches of banana conducted in Guanambi, BA, with genotypes Prata-Ană (AAB) and 'BRS Platina' (AAAB), planted in a spacing of 3.0 m x 2.5 m, under irrigation. Measurements of vegetative and yield characteristics were performed in 98 'Prata-Ană' plants and 96 BRS Platina one at random We determined the means and standard deviations, correlation and regression to estimate the harvest prediction models. We estimated simple linear regression models, considering as independent variables those with the highest correlation coefficients with bunch and hands masses. They were also adjusted multiple linear regression models by backward elimination using as independent variables yield characteristics with higher coefficients of correlation. The models of simple linear regression esteem with lesser accuracy the masses of bunches (MCA) and hands (MPE) based on pseudostem perimeter measured at level of the soil for 'Prata-Ana' and 'BRS Platina' bananas. The models of multiple linear regression esteem with adequate accuracy the masses of bunches and hands from the independent variables number of fruits and fruit mass of the fourth hand for 'Prata-Ana' and number of fruits and fruit mass of the fifth hand for 'BRS Platina'. The models of simple linear regression make possible prediction of the masses of bunches and hands, in function of the number of hands (NPE), with better accuracy, for both the genotypes and with antecedence of at least 120 days to harvest. Being the following equations for 'Prata-Anã': MCA = -10.0485 + 3.07985**NPE; r² = 0.99, and, MPE = -12.3804 + 3.55401**NPE; r² = 0.99 For 'BRS Platina', the equations were MCA = -1.36638 + 3.01573**NPE; r²= 0.97 and MPE = 2.73380+ 2.87825**NPE; r²= 0.97 The values of determination coefficients for the adjusted models assure the consistency of the regressions to estimate production of banana trees type 'Prata'.