Utilização de Python na estimativa de produção da cana-de açúcar a partir de índices de vegetação

Abstract

The sugarcane crop holds significant economic importance in Brazil, especially in the producing regions of the Northeast. In this context, the use of agricultural monitoring technologies, such as remote sensing and statistical modeling, has proven essential for accurate and efficient yield estimation. This study aimed to estimate sugarcane production based on the application of the vegetation indices NDVI, SAVI, and EVI, obtained from multispectral images of the Sentinel-2 satellite, using an automated processing workflow in Python. The study area comprised a rainfed field of approximately 60 hectares located in the municipality of Maxaranguape, in the state of Rio Grande do Norte. The images were acquired between 53 and 121 days before harvest, a period during which the crop is in an advanced stage of vegetative development. The processing was carried out using the Rasterio, GeoPandas, and Scikit-learn libraries. The spectral indices were used as independent variables in a multiple linear regression model, which yielded a coefficient of determination (R²) of 0.8695 and a root mean square error (RMSE) of 147.65 t, approximately 2.46 t/ha. The results indicated a strong correlation between the vegetation indices and the production observed in the field, highlighting the performance of the 2020 and 2022 harvests. Factors such as the image acquisition window, water availability, and operational aspects like harvest delay directly influenced the accuracy of the estimates. The use of vegetation indices integrated with computational tools offers a viable, replicable, and low-cost alternative for agricultural production monitoring.