Estimativa de parâmetros de vegetação e de qualidade de água usando sensoriamento remoto

Abstract

Remote sensing is a technique that allows obtaining information about different targets without physical contact. This device is potentially used in agricultural and environmental activities, promoting more efficient monitoring and management of activities. Monitoring the nutritional status of plants is essential for the early identification of nutrients that can limit the growth and production of crops, as well as monitoring water resources is of paramount importance for decision making and environmental management. Both monitoring are commonly performed through laboratory analysis, through procedures that take time, maintenance cost and high waste production. Thus, the present work aimed to quantify, by means of reflectance spectroscopy, the contents of leaf nutrients of the cowpea crop and the limnological variables chlorophyll-a, total suspended solids and transparency, as an alternative to laboratory analyses. The cowpea crop was analyzed in three phenological stages: V4, R6 and R9. Leaf samples were collected for radiometric readings and quantification by wet nitro-perchloric digestion. The acquisition of water samples was carried out in the Caxitoré, General Sampaio and Pereira de Miranda reservoirs, where the collection was carried out at 3, 4 and 5 points, respectively, in parallel with the water collection, hyperspectral data were obtained by in situ radiometry. The analyzes of plant tissues and water samples were performed in the laboratory and the acquisition of radiometric data was carried out using the FieldSpec 3 Hi-Res spectroradiometer. Three models were built for each variable, based on: simple correlation, 2D correlation (band ratio) and Partial Least Squares Regression (PLSR). The most significant spectral variables for the simple correlation and 2D correlation models were selected by the highest value of r, while for the construction of the PLSR models, the Stepwise method was used. 70 and 30% of the data were used for the construction and validation of the prediction models, respectively. In order to validate the results, statistical metrics were applied: determination coefficients, RMSE and RPD. The PLSR predictive models presented better performances when compared to the simple and 2D correlation, both for leaf nutrients and for limnological variables, presenting adjusted R² of 0.97, 0.23, 025 and 0.23 and RPD = 1.23, 0.25, 6.09 and 2.38 for the predictions of P, K, Ca, and Zn, respectively and adjusted R² of 0.23, 025 and 0.23 and RPD = 0.25, 6.09 and 2.38 for chlorophyll-a, SST and transparency, respectively. It was possible to observe that the most significant wavelengths in the prediction of leaf nutrients are in the visible region, while for the limnological variables they are in the near infrared (NIR) regions.