Crop phenology drives remote sensing prediction accuracy of sugarcane yield and quality in the Brazilian Cerrado

Crop phenology influences the relationship between remotely sensed signals and agronomic traits; however, its effect on the prediction of sugarcane yield and technological quality remains insufficiently understood under tropical production conditions. This study evaluated the influence of crop growth stage on remote sensing-based estimates of sugarcane yield and total recoverable sugar (TRS) in the Brazilian Cerrado. Field data were collected from 51 georeferenced sampling points distributed across five commercial sugarcane fields totaling 98.1 ha and cultivated with the CTC-04 variety in Campo Florido, Minas Gerais, Brazil. Vegetation indices derived from Sentinel-2 imagery, including NDVI, EVI, NDRE, GNDVI, and VARI, as well as RGB- and ALOS/PALSAR-2-derived metrics, were evaluated at vegetative and pre-harvest stages. Multiple linear regression models were developed to estimate total cane yield (TCH), net cane yield after removal of leaves and apical stalk portions (TCH-L), and total recoverable sugar (TRS). Total cane yield and TCH-L were more accurately estimated using optical vegetation indices, with the Normalized Difference Vegetation Index (NDVI) providing the most consistent performance and explaining up to 60% of TCH variability, with mean absolute percentage error (MAPE) values below 20%. In contrast, TRS showed a distinct phenological response, achieving MAPE values below 5%, with its best estimates during the pre-harvest stage when the Normalized Difference Red Edge Index (NDRE) and NDVI were combined.