Influences of sampling design and model selection on predictions of chemical compounds in Petroferric formations in the Brazilian Amazon

Authors: BRUNO ROGRIDUES, NIRIELE - Federal Rural University Of Rio De Janeiro; BARBOSA ROCCO, THERESA - Federal Rural University Of Rio De Janeiro; PINHEIRO, HELENA - Federal Rural University Of Rio De Janeiro; MANCINI, MARCELO - Federal University Of Lavras; Read, Quentin; Blackstock, Joshua; WINZELER, HANS - University Of Texas At Arlington; MILLER, DAVID - University Of Arkansas; Owens, Phillip; Libohova, Zamir

Submitted to: Remote Sensing
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/27/2025
Publication Date: 5/6/2025
Citation: Bruno Rogridues, N., Barbosa Rocco, T., Pinheiro, H.S., Mancini, M., Read, Q.D., Blackstock, J.M., Winzeler, H.E., Miller, D., Owens, P.R., Libohova, Z. 2025. Influences of sampling design and model selection on predictions of chemical compounds in Petroferric formations in the Brazilian Amazon. Remote Sensing. https://doi.org/10.3390/rs17091644.
DOI: https://doi.org/10.3390/rs17091644

Interpretive Summary: Rare earth elements are widely used by medical industry for instruments used for diagnosing several human medical conditions and diseases. However, in many instances the rich formations with rate elements occur in isolated areas with limited access. Thus an accurate assessment of their occurrence and amount is crucial for saving cost and the environment during extraction. By using several machine learning algorithms this study tested different sampling designs and identified the ones that deliver accurate maps of the distribution of these rare earth elements. Finding from these study can be implemented in other areas for multiple rare earth elements thus reducing the footprint of extraction for rural areas and local farmers.

Technical Abstract: Morro de Seis Lagos a region in the Brazilian Amazon hosts a rare (less than 1%) formation of siderite carbonatites, considered one of the world's largest Niobium reserves. This highly weathered geological and pedological occurrence makes the site ideal for studying pedogenetic process of lateralization and the spatial variability of the chemical elements. The aim of this study was to investigate the influence of various sampling combinations (scenarios) derived from three sampling designs on spatial predictions of chemical compounds (Al2O3, Fe2O3, MnO, Nb2O5, TiO2, and SiO2), using multiple machine learning algorithms combined with remotely sensed imagery. The dataset comprised of 341 samples from the Geological Survey of Brazil (CPRM). Covariates included remotely sensed data collected from Sentinel-2 MSI, Sentinel-1A, and Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), and topographic attributes calculated from a 20-m digital elevation model derived from hydrologic data (HC-DEM). Machine learning algorithms (Generalized Linear Models with Elastic Net Regularization (GLMNET), Nearest Neighbors (KNN), Neural Network (NNET), Random Forest (RF) and Support Vector Machine (SVMRadial) were used in combination with covariates and measured elements at point locations to spatially map the concentration of these chemical elements. The optimal covariates for modeling were selected using Recursive Feature Elimination (RFE), processing 10 runs for each chemical element. RF, SVMRadial, and KNN models performed better followed by the models from the neural network group (NNET). The different sampling scenarios were not significantly different based on root mean square error; however significant differences were observed for the coefficient of determination across all models. The terrain attributes were significantly more influential on the spatial predictions of the elements contained in laterites than the remote sensing spectral indexes, likely due to the underlying spatial structure of the two formations (laterite and talus) occurring at different elevations.