Análise da fenologia e produtividade da soja utilizando índices de vegetação em diferentes resoluções espaciais

Abstract

The objective was to identify vegetation indices with the potential to identify phenological stages and estimate soybean yield using data from different sensor systems. The research was conducted in Balsas, Maranhão, Brazil, in a center-pivot irrigated area, with weekly assessments of phenology and Leaf Area Index (LAI). Spectral data were collected from the study area using the MSI and WFI sensors onboard the Sentinel-2 and Amazônia-1 satellites, respectively. The data were processed to calculate the NDVI, NDWI, SAVI, VARI, Green VI, and GLI indices. The efficiency of identifying phenological stages was determined using Anderson's discriminant analysis and the Random Forest machine learning algorithm. Yield estimation was performed using path analysis (a multivariate technique), with pre-filtering based on Pearson correlation represented by correlation networks. No single vegetation index from the sensor systems showed 100% effectiveness in identifying all phenological stages; however, specific RGB and NIR indices can be used to identify key stages, such as flowering (R1 and R2), the end of pod development (R4), and grain development within pods (R4 to R5.2). Vegetation indices—whether used individually or in RGB and NIR groups—are not efficient for identifying phenological stages on their own; using the indices jointly with the Random Forest algorithm is the most efficient method for this purpose. Furthermore, the Sentinel-2 sensor system outperforms the Amazônia-1 system in identifying soybean phenological stages. Regarding yield estimation, the SAVI and NDVI (infrared) indices enabled satisfactory estimation of soybean yield at the V5 and R2 stages, respectively, using the Amazonia-1 sensor system, while the VARI (RGB) index did so at the R2 and R5.2 stages using the Sentinel-2 sensor system. The spatial resolutions employed in this study were not a limiting factor for soybean yield estimation. Vegetation indices derived from orbital sensors are effective for rapidly and cost-efficiently identifying soybean phenological stages