SENSORES DE REFLETÂNCIA ESPECTRAL E DESEMPENHO DA CULTURA DO TRIGO EM RESPOSTA À ADUBAÇÃO NITROGENADA EM PLANTIO DIRETO

Abstract

No-till systems with diversified crop rotations have stood out of the most effective strategies to improve the sustainability of farming in tropical and subtropical regions. Wheat (Triticum aestivum L.) is one of the most important crops used in this rotation during the autumn-winter season. Nitrogen (N) is uptake in larger amounts by plants, it is essential for the structure and functions in the cell, for all enzymatic reactions and is part of the chlorophyll molecules. Nitrogen fertilizers represent a significant part of the costs of production and due to the dynamics of N in soil, losses of N occur and cause economic and environmental damages. In the same agricultural area may exist changing demands for this nutrient. The attributes of the plant commonly used as indicators of N are NO3- content in stem, leaf chlorophyll content, the intensity of the green color and the N foliar content, dry biomass, and the extraction of N by plants. Lower levels of N can cause chlorophyll deficiency that is recognized by whitish or pale foliar coloration, and this changing in plant color can be identified using remote sensing techniques. This study aimed to evaluate the correlations between spectral reflectance data obtained by commercial ground sensors (Clorofilog 1030, GreenSeeker, and Crop Circle ACS-470) and attributes of wheat crop in response to N rates in top dressing under a no-till system. The efficiency of the sensors was evaluated in two ways: (i) by classical statistical methods, and (ii) through the application of Artificial Neural Networks, a machine learning technique. For the use of Artificial Neural Networks, this study compared the performance of the algorithms Resilient Propagation and Backpropagation. Because wheat plants exhibited adequate nutritional status, even without N application in top dressing, Clorofilog 1030 readings were not sensitive to variations of N rates. Thus, this sensor also did not correlate significantly with the N foliar content, dry biomass, and the extraction of N by wheat plants. The indices obtained by reflectance sensors Crop Circle and GreenSeeker had close correlation with the rates of N in top dressing, dry biomass, and the extraction of N by wheat plants. The Crop Circle and GreenSeeker sensors showed weaker correlation with the N content in leaves, and especially with the wheat grain yield. In this way, it was evident that grain yield has not followed the dry biomass production when high wheat grain yields were obtained. The correlation coefficients obtained by the Resilient Propagation and Backpropagation algorithms were similar to those found by statistical analysis. The Artificial Neural Networks technique had satisfactory behavior similar to classical statistical methods.