Forecasting total recoverable sugar index and sugarcane production by using multiple regression models

Sugarcane must be harvested at the time of maturity, known as the Period of Industrial Utilization (PIU). The PIU analysis is expensive and is performed in a laboratory by measuring an index defined as Total Recoverable Sugar (TRS). The fact is that harvesting is the most expensive stage in sugarcane production and decision-making in this segment depends on TRS level estimates. However, forecasting models aimed at replacing laboratory analyses do not meet the reality regarding the estimation of the TRS index. There is a great demand for tools capable of estimating the index and/or the factors that affect TRS. In this context, this article presents a case study whose objective is to apply statistical models to estimate the TRS index in the sugarcane production of a mill in the interior of São Paulo, Brazil. Variance Analysis (ANOVA) with one classification and Multiple Linear Regression (MLR) models are applied by using Minitab®. These models are based on covariates related to the TRS index to estimate the productivity of 48,151 plots from the 2016/2017 to the 2022/2023 harvests. It is shown that the adjusted models identify the most important covariates (5% significance level) that affect productivity and the TRS index. The accuracy is satisfactory for all covariates of the adjusted MLR model and for the coefficients that measure the proportion of data variability for productivity (76%) and the TRS index (55%). This article brings important contributions to the sugar and ethanol industries worldwide and in Brazil.