Previsão de safras de cultivos anuais: Estimativa com índices espectrais de vegetação

Abstract

Brazil is the world's leading producer of soybean and third in corn production. Orbital remote sensing has broadened the range of techniques and methodologies that assist in crop monitoring. Vegetation indices can be used, from mathematical models, to estimate the crop productivity. The objective of this work was to develop an empirical model for the prediction of soybean and corn harvest crops, using spectral vegetation indices - IV - with medium resolution orbital data. The study was carried out with soybean and corn crops data grown on the Porto Alegre Farm, municipality of Serra do Ramalho, West Bahia, in the 2018/2019 and 2019/2020 seasons. Surface reflectance images from the Landsat 8 satellite were applied to compute NDVI and SAVI vegetation indices, and by means of the Google Earth Engine, the maximum value of these indices for the crops in each field plot was determined. This data was applied with the field yield values to generate an empirical equation to estimate the harvest for each crop. Vegetation index values were homogeneous within the plots, however low, with a maximum NDVI around 0.55. The regression that best fitted the data was quadratic, with R² < 0.5; however, Tukey's test indicated that the averages of estimate and measured production were statistically equal. Probably, the few sampled data contributed to the unsatisfactory model performance. Other aspects (phytosanitary, nutritional and meteorological variables) may explain the fact that, even with an equal maximum IV, the plots had different productivity. The use of a single image does not translate other aspects that contribute to the production, especially in the final crop stage.