Caracterização espectral de trigo em lavoura comercial e estimativas biofísicas e qualidade de grãos baseada em quimiometria

Abstract

With population growth, it is crucial to increase agricultural production in order to ensure food and energy security. As a result, understanding factors that affect crop yield and quality will be essential to face future challenges regarding agricultural fluctuations, mainly caused by global climate change, water demand and soil limitations. The total protein contained in cultivated grains is imperative for the final use of cereal flour, however, total protein measurements are time consuming and high-cost analyses. For this reason, the operation of remote sensing techniques to estimate protein content and grain yield is extremely important for the agricultural producer, in addition to being a non-destructive technique, which makes it possible to obtain prior information on the grain quality with the crop still in the field and in a spatialized way. Therefore, the objective of this study was to estimate biophysical characteristics and grain quality of wheat crops based on the leaf spectrum acquired by hyperspectral sensor with chemometric techniques, such as partial least squares regression (PLSR). The experiment was carried out on a rural property in the municipality of Céu Azul - PR, where meteorological stations are installed for climate data collection. Field analyses were carried out, such as leaf spectroscopy with hyperspectral sensor and laboratory analyses, including leaf pigment content (chlorophyll and carotenoids), protein content and grain hectoliter weight and yield. When estimating chlorophyll A and chlorophyll B, it was obtained mean square error (MSE) between 1.9 and 27.66 µg cm-2 and coefficient of determination (R²) from -0.12 to 0.1, demonstrating lack of model adjustment for the estimation of these parameters. For the estimation of carotenoids, MSE from 0.59 to 1.93 and R² from -0.2 to 0.46 were obtained, demonstrating, likewise, lack of model adjustment for the estimation of these parameters. Regarding the estimates of grain quality parameters, the models also showed a lack of adjustment, given that, for the grain protein content, the MSE ranged from 0.12 to 0.97% with R² between -0.09 and 0.39, evidencing that the phenological stage of booting displayed lower MSE and higher R² for both crops under study. In the estimation of hectoliter weight (PHEC), the MSE values varied from 1 to 1.53, and R² from - 0.02 to 0.22. The lack of model adjustment was due to the low amplitude of variation of the sample database (pigments and quality parameters). The utilization of data from commercial fields is not ideal for the elaboration of PLSR models, requiring conditions that allow a greater range of variation of input data in order to promote the best fit of the model.