Desenvolvimento de equipamento óptico para avaliação nutricional em plantas: aplicação na cultura do milho

Abstract

The search for technologies that promote more sustainable and efficient practices is an increasingly important priority on the global agricultural scene. With this in mind, the use of spectroradiometry techniques to assess nitrogen concentration in maize crops has proved to be a promising alternative for conducting precision agriculture practices. In addition, the development of new instruments with optical sensors for agricultural applications has become increasingly important given the challenges of maintaining the sustainability of production systems. With this in mind, this work sought to improve elements of the use of MSPAT (Multispectral Soil Plant Analysis Tools) to determine reflectance. The improvements involved the development of a reference plate with sintered barium sulphate, new approaches using programming techniques, electronics and 3D printing. In addition, the performance of the optical instruments used to estimate leaf nitrogen in the maize crop was also assessed. With this in mind, an experiment was carried out in the experimental area of the Agricultural Electronics and Mechanisation Laboratory (LEMA), located at the Federal University of Ceará (UFC), with AG-1051 maize planted under treatments N0, N60, N90, N120, N150 and N180 (0, 60, 90, 120, 150 and 180 kg.ha-1 of N, respectively) with four replications, in two crop cycles and a Completely Randomized Design (CRD). The assessments took place during the V5, V10 and R2 phenological stages and were carried out using the MSPAT, SPAD and FieldSpec PRO FR 3 equipment. The spectroradiometer was used under the conditions provided by the dark-room of the UFC Geoprocessing Laboratory. The samples were then prepared for nitrogen (N) determination according to the methodology proposed by Kjeldah. In addition to leaf nitrogen, morphological parameters were also assessed throughout plant development, production, biomass and dry matter. The spectral indices (NRI, Normalise Ratio Index) were then correlated with the leaf N data and the individual bands were evaluated using Pearson's coefficient (r). Linear regression was then carried out (p-value < 0.01) with the NRI's to select the models, by optical instrument, that showed the best coefficient of determination (R²) with the leaf N data sets: i) at each stage of development, for the two crop cycles and ii) with the entire data set. Cross-validation (k-fold) was then carried out to assess the error parameters RMSE, MAE and adjusted coefficient of determination (R²adj.). The results for the best predictive models reveal different patterns for the selected bands between the data sets, as well as low generalisation capacity. However, it was possible to validate relevant models with MSPAT, SPAD and FieldSpec, which showed an R²adj. of 0.7871; 0.6959; 0.7199 and RMSE of 0.0425; 4.47; 0.0214 g.kg-1; respectively. From the model that showed the best performance with MSPAT, when using the NRI with the 900 and 560nm bands, the application in an agricultural area provided an RMSE of 2.73 g.kg-1 and MAE of 2.47 g.kg-1. However, it is clear that the use of new technologies has great potential for assessing leaf N in maize crops.