Uso do sensoriamento remoto para diagnóstico nutricional na cultura do milho irrigado

• Ano de defesa: 2018
• Autor(a) principal: Campelo, David de Holanda
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Palavras-chave em Português: Agricultura de precisão; Hiperespectral; Índice de nutrição; PSLR; Zea mays
• Link de acesso: http://www.repositorio.ufc.br/handle/riufc/34754

The estimation of status plants nutritional from remote sensing can contribute to the monitoring and optimization of the management of fertilizers in the crops, providing both agronomic and environmental benefits. The concentration of leaf nitrogen (N) is an important indicator of crop health, playing a key role in the photosynthetic metabolism and plant nutrition. Thus, the quantification of N is relevant for predicting productivity and improving the management of fertilizer application. Hyperspectral remote sensing is a timely, promising technique with high application potential in agriculture at various levels of spatial coverage to estimate, quantify and monitor biophysical and biochemical attributes of vegetation. This study aimed to evaluate the potential application of SR development data methodological strategies to identify nutritional deficiency and quantify the concentration of leaf nitrogen (N) in irrigated maize (Zea mays L.). For that, two experimental strategies were executed. In the first part of the study, an experiment was set up with a randomized block design with subdivided plots and four replications. The irrigation treatments applied in the plots were composed of four levels: 80, 90, 100 and 110% of the water requirement, based on the soil field capacity and the nitrogen levels, distributed in the subplots at levels 0, 60, 120 and 180 kg ha-1. The production components and water use efficiency (US) and N (EUN) were evaluated, spectral readings were performed, and the concentration of leaf nitrogen at the V8, R1 and R3 stages of the crop at the field and laboratory levels was determined. Two cultivation cycles were carried out in the years 2015 and 2016. The results indicated that the irrigation slides and N doses influenced plant height and leaf area index. The maximum yield was reached the corresponding blade to 505 mm and 180 kg ha-1 C, in a yield of 17819.5 kg ha-1. The highest EUA was obtained in irrigation regimes reduced to 80 and 90% of the water requirement, while the highest EUN (67.5 kg kg-1) was reached with the as the 555.7 mm blade and level 113, 3 kg ha-1 of N. Regarding the spectral responses, the strategies used to analyze the reflectance profiles were derivative analysis (AD) and principal component analysis (PCA). The results evidenced sensitive regions of foliar N concentrations located mainly in the visible range in the range of 400 nm to 700 nm and near infrared between 800 nm and 1300 nm. The level of data collection was an important factor that influenced the spectral response as a function of fertilization treatments. In the laboratory, the regions between 450 nm-750 nm, are more important and at the field level the best response can be observed between 800 nm and 1300 nm. The first-order AD identified greater throne variations of two specific regions 470 nm at 550 nm and 720 nm at 750 nm. The second order AD, in turn, was able to show the range coincident with the response of chlorophyll pigments between 700 nm and 725 nm. ACP showed that about 90% and 95.12% of the total variation in spectra can be explained using three major components. For field data (canopy) the reflecting power over the entire spectrum dominated PC1. For laboratory data (sheet), the same PC1 is influenced mainly in the red border range (around 700 nm) and throughout the infrared and part of the SWIR (750 nm at 2500 nm). The PC2 for the datasets is distinguished mainly between the bands 750 nm to 1300 nm, by factorial charges, positive for leaf level and negative for canopy. PC3 is influenced in the visible, with maximum peaks around 550 nm and 750 nm in part of the SWIR at 1800 nm. With the same spectroscopy data, the ability to generate predictive models that were developed using linear regressions of individual bands, normalized nitrogen index (NDNI) and partial least squares regression (PLSR) were also investigated. The leaf N was precisely quantified, obtaining better results for laboratory PLSR models in the R1 stage (R² = 0.82 and RMSEP = 0.190). Using field-level data the best result was obtained by the NDNI (762 nm and 684 nm) for the R3 stage (R² = 0.78 and RMSE = 0.200). Modeling was influenced by the phenological stages and level of data collection, which reinforces the importance of adjusting predictive models as a function of the different stages of crop growth. In the second part of the study, the potential of the use of hyperspectral images obtained with the ProSpecTIR-SV airborne sensor to estimate leaf N in corn at different stages of growth was investigated, in order to evaluate the potential of this tool in the management of fertilization. Samples of leaf N of the canopy of plants were collected in different fields of maize in the vegetative (1m) and reproductive (2m) stages during the first half of 2015. The stepwise forward spectral band selection procedure was used in conjunction with the partial least squares regression (PLSR) for the development of the models. Data from the reproductive stage showed a better performance in leaf N estimation (R² = 0.844, RMSECV = 1.0114 and RPD = 2.388). The mixed model (1m + 2m) was less accurate than the invidious models. The potential of the images for the creation of N fertility maps was accurately confirmed with RMSECV ranging from 1,014 to 2,082 g Kg-1 N. Finally, band selection along with PLSR is promising for leaf N detection in the crop of maize.