Sensoriamento remoto hiperespectral na estimativa da granulometria de horizontes superficiais de solos

• Ano de defesa: 2020
• Autor(a) principal: Almeida, Eurileny Lucas de
• Orientador(a): Teixeira, Adunias dos Santos
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: por
• Área do conhecimento CNPq: CNPQ::CIENCIAS AGRARIAS::AGRONOMIA::CIENCIA DO SOLO
• Link de acesso: http://repositorio.ufc.br/handle/riufc/78944

The general objective of this research was to estimate soil texture, through its spectral behavior obtained by laboratory spectroradiometry and airborne sensors, with and without the interference of Organic Matter (OM) in different soils in the state of Ceará. The study was carried out in two different areas of the state of Ceará, in the municipalities of Morada Nova (A1) and Limoeiro do Norte (A2). 233 deformed soil samples were collected, from 0 to 10 cm deep. For each soil sample, particle size, OM and spectral data were obtained using the ProSpecTir-VS airborne sensor and the FieldSpec Pro FR 3 spectroradiometer sensor in the laboratory. To obtain spectral data in the laboratory, Oven-Dried Fine Earth at 45ºC (TFSE) and Fine Earth Without Organic Matter (TSOM) were used. A descriptive analysis of particle size and spectral data was carried out and Pearson's correlation was obtained between the sand, silt and clay contents and the reflectance of the soil without OM. In the analysis of spectral data acquired in the laboratory from soil samples, with and without Organic Matter, Principal Component Analysis - PCA was also applied. To estimate soil texture, all possible band relationships of the two sensors were tested in search of a Normalized Difference Index (NDI). In addition to reflectance, the spectral data were also analyzed in transformed form: Savitzky-Golay smoothing, 1st derivative and normalized. Partial Least Squares Regression (PLSR) was applied using all spectral data and after band selection. For model calibration, 70% of the soil samples were used and 30% for validation. The analyzes were carried out using the total set of data and separated by region (A1 and A2). Thus, it was observed that the soils in the A2 region (Irrigated Perimeter Jaguaribe Apodi) are more clayey than the soils in A1 (Irrigated Perimeter Morada Nova), with the latter having a predominance of silty and sandy soils. The best correlation results were from the proximal sensor, FieldSpec Pro FR 3, for clay, with a strong correlation of -0.74 and -0.71 for the complete sample and for the A1 region, respectively. The wavelengths chosen to construct the NDI were 2133 and 2335 nm, with a coefficient of determination (R²) of 0.67. The best validation results, using PLSR, were from the laboratory sensor with data in first derivative, with adjusted R² of 0.77 and 0.79 for clay using all data and for sand with A2 data, respectively. Regarding texture estimation for soil without OM, the best PLRS results were for sand with all normalized data (adjusted R² = 0.75) and for A2 (adjusted R² = 0.75), using spectral data without transformation. It is concluded from this work that laboratory spectral data (FieldSpec) were more efficient in estimating the textural attributes of soils than airborne sensor data (SpecTIR-VS), especially when using a data set with different soils and regions. . When comparing the predictive models, using the spectral behavior of soil samples with and without organic matter, it is possible to notice the improvement in the estimation of sand and clay contents, after removing the OM.