Banca de DEFESA: CINTHIA MARIA CORDEIRO ATANAZIO CRUZ SILVA

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : CINTHIA MARIA CORDEIRO ATANAZIO CRUZ SILVA
DATE: 11/08/2023
TIME: 09:00
LOCAL: Auditório do PPG em Ciência do Solo
TITLE:

Near infrared applied to the mapping of physical attributes and chemical elements in soils combining mathematical models and pre-processing techniques

KEY WORDS:

diffuse reflectance spectroscopy. Environmental quality. Chemical elements.

PAGES: 98
BIG AREA: Ciências Agrárias
AREA: Agronomia
SUBÁREA: Ciência do Solo
SPECIALTY: Química do Solo
SUMMARY:

The soil is an open system formed by organic and inorganic constituents that can be evaluated by diffuse reflectance spectroscopy. Thus, the present study aimed at testing different combinations of preprocessing and calibration methods for the prediction of chemical elements (Ba, Cd, Co, Cr, Cu, Fe, Mn, Ni, Pb, Se, Sb, V, Zn, Sc, La, Ce, Pr, Nd, Sm, Eu, Y, Gd, Tb, Dy, Er, Ho, Yb, Lu), group of elements (LREEs, HREEs, LREEs/HREEs, REEs), granulometric attributes, pH, cation exchange capacity and total organic carbon from the spectra obtained in the near-infrared region (NIR) as pedoindicator attributes. The soil samples were select to representa the geological and pedological diversity of the northeast Brazil. We hypothesized that (i) regarding the prediction of chemical elements in soils derived from different parent materials, the choice of the mathematical model is more important than the choice of pre-processes and (ii) under the same geological, pedological and climatic context, mathematical models of soils in the NIR range from other areas can be used for the prediction of chemical elements in areas where only spectral signatures exist. The states of PE, PB, and RN were select for this research because they represent the pedological and geological diversity of Brazil, which increases the applicability and scope of the results. All 13 soil orders in the Brazilian System of Soil Classification occur in the study area. Measurements in the NIR length range (1000-2500 nm) were performed in a Fourier transform FTIR/ NIR spectometer (Frontier/Perkin Elmer), coupled with a near Infrared reflectance accessory (NIRA). The Random Forest (RF) model had the best performance. Furthermore, it was also found that the choice of the model was more relevant than the choice of pre-processing. The better prediction of REEs compared toheavy metals can be observed by the higher values of the predicted residual deviation (RPD) and the ratio between the performance and the interquartile distance (RPIQ) and the lower values of the associated errors. Future studies must to further explore the combination of models with different computational natures rather than testing various preprocessing techniques with single models. The predicted values for geographic regions where only spectral signatures existed showed moderate spatial dependence, except for Pr, Sm, and Tb, which showed strong spatial dependence. This reinforces the quality of the predicted maps for REEs, which are essential for identifying areas susceptible to environmental impacts - an important step in establishing environmental policies for the protection of human health and the environment

COMMITTEE MEMBERS:
Presidente - CLISTENES WILLIAMS ARAUJO DO NASCIMENTO
Interno - EDIVAN RODRIGUES DE SOUZA
Externa à Instituição - MARIA EUGENIA ORTIZ ESCOBAR - UFC
Externo à Instituição - RENNAN CABRAL NASCIMENTO
Externo à Instituição - RONNY SOBREIRA BARBOSA - UFPI