Banca de DEFESA: NIRIELE BRUNO RODRIGUES

Uma banca de DEFESA de DOUTORADO foi cadastrada pelo programa.
STUDENT : NIRIELE BRUNO RODRIGUES
DATE: 31/07/2025
TIME: 09:00
LOCAL: VIDEOCONFERÊNCIA
TITLE:

Digital mapping in mineralized petroferric formations: Case study - Morro dos Seis Lagos in the Brazilian Amazon

KEY WORDS:

Pedometry. Machine learning; Poorly-Accessible Areas. Pedogenetic processes. Ferrocarbonatites.

PAGES: 124
BIG AREA: Ciências Agrárias
AREA: Agronomia
SUBÁREA: Ciência do Solo
SUMMARY:

The combination of remote sensing and topographic data associated with Machine Learning (ML) models, especially for mapping digital geological and pedological formations, has contributed to the identification of areas with economic potential for mineral prospecting. However, the pedological aspects of these formations are still little explored. The carbonatite formations, in particular, are rare and the Brazilian Amazon is home to a particular mineralization of siderite carbonatite, representing the largest known deposit of niobium (Nb). This formation contains a thick lateritic crust (>200 m), where the carbonation processes of the siderite have produced a goethite/hematite crust. The objectives of this research were: to investigate the lateritization process in rare siderite carbonatite (ferrocarbonatites) by evaluating: spatial variability; the effect of the design of the sample mesh on the elements Al2O3, Fe2O3, MnO, Nb2O5 and TiO2; the distribution of geological materials (subclasses); and weathering indices (Index of Lateritisation - IOL, Chemical Index of Alteration - CIA, and the Chemical Index of Weathering - CIW). Machine learning was used to combine morphometric covariates with remote sensing images. The input data set includes geochemical data from 341 samples (soil, sediment and rock materials), combined with morphometric covariates and spectral indices obtained from remote sensing data. This information was extracted from bands of the Sentinel-1A SAR, Sentinel-2A and Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) sensors, as well as topographic attributes derived from a hydrologically conditioned digital elevation model (MDE-HC), with a resolution of 20 meters. The data collected was modeled using machine learning algorithms. The most important covariates for each mineral compound and for each model were selected using the Recursive Feature Elimination (RFE) algorithm. The models with the best predictive performance were Random Forest (RF), followed by K-Nearest Neighbors (KNN). The ranking of importance provided by the RFE indicated that, initially, terrain attributes exert a more significant influence on the variability of the elements contained in the laterites than remote sensing spectral indices. In this context, it was found that the characteristics of the local relief play a fundamental role in understanding the spatial variation of mineral compounds, given the greater influence of morphometric and SAR covariates in predicting the different elements and compounds.

COMMITTEE MEMBERS:
Externo à Instituição - CARLOS ERNESTO GONÇALVES REYNAUD SCHAEFER - UFV
Externo ao Programa - 1580057 - FRANCISCO JOSE DA SILVA - nullExterno à Instituição - GUSTAVO DE MATTOS VASQUES - EMBRAPA
Presidente - 2223668 - HELENA SARAIVA KOENOW PINHEIRO
Interna - 387335 - LUCIA HELENA CUNHA DOS ANJOS