Uso de sensoriamento remoto hiperespectral na caracterização da cultura do algodoeiro (Gossypium hirsutum L.)

Abstract

Remote Sensing provides technologies and knowledge useful for increasing agricultural yields and does so by generating solutions that optimize the management of these nutrients and the perspective of the producer. Potassium is one of the most important biochemical components of plant organic matter, so estimating its components can help monitor metabolism processes and plant health. In this context, the objective was to evaluate the agronomic parameters and, then, the reflectance factors in the cotton crop of cultivar BRS 293. The research was carried out in a semi-protected environment located in the experimental area at the Hydraulics and Irrigation Laboratory of the Department of Agricultural Engineering of the Federal University of Ceará - Campus do Pici, in the city of Fortaleza, CE. A total of 166 plants were cultivated in low density polyethylene pots, filled with arisk and subjected to a completely randomized experimental design, whose treatments corresponded to four levels of nitrogen (N) and potassium (K) (50%, 75% 100% and 125% of the nutritional demand), where all other elements were satisfied, with twenty repetitions and kept under the same daily irrigation depth. The highest yields were reached in 264.67 kg.ha-1 and 348.81 kg.ha-1 of feathers under the doses of 73.0 kg.ha-1 of K (100%) and 86.25 kg.ha -1 N (125%), respectively. The physical quality of the fibers of each treatment was also verified, emphasizing the doses N4 (125%) and K3 (100%) as the best micronaire indices, reliability and reflectance. Fertilization levels of both data groups strongly interfered with the spectral profiles of cotton plants. Variations in the derived curves have individualized strategic evaluation points for both treatments, namely: wavelengths 510, 690, 997, 1152, 1390 and 1880 nm of sharp ascendancy or descent and characteristic inflection points at 550, 590, 667, 715, 730, 1006, 1130 and 1380 nm. Second-order derivative analysis and PCA were more efficient in identifying variations between treatments in the following wavelength ranges: i) from visible (380 to 750nm) for nitrogen; and ii) in the plant water absorption ranges (1400 and 1800 nm) for potassium. The performance of the PLSR models on cotton leaf chemometrics was analyzed as a function of their adjusted coefficient of determination, root mean square error and residual prediction deviation. The validation results indicated that the PLSR method is useful in the elaboration of a predictive model capable of capturing 82.0% of the variation in leaf potassium concentration, with RMSE of [3.74] and RPD of [1.61], presenting a good estimation ability.