Evaluation of alternatives for simplifying the estimation processes of chemical and granulometric variables

Iron ore resource modeling is challenging due to the multivariate characteristics of the deposit, particularly the need to integrate chemical composition and granulometric size fractions while ensuring mass and stoichiometric balance. This research investigates the potential to simplify the estimation of Brazilian iron ore resources while preserving modeling accuracy. Ordinary kriging (OK), intrinsic coregionalization model (ICM), and linear coregionalization model (LCM) were applied to chemical (Fe, SiO2, P, Al2O3, Mn, PF) and granulometric (G1–G4) data from drill hole samples. All three approaches yielded robust and internally consistent block models. Validation using swath plots, grade-tonnage curves, and misclassification rates showed strong agreement between models, though discrepancies increased at upper grade ranges, especially for Mn, PF, and P. The OK method demonstrated superior adaptability to local spatial continuity, while ICM and LCM maintained inter-variable correlation structures more effectively. Mass and stoichiometric balance checks validated the models, with closure ranges of 97%–103% for chemistry and 99%–101% for granulometry. The results suggest that model selection should align with project priorities—favoring OK for spatial resolution and ICM/LCM when correlation fidelity is paramount. The study's novelty lies in demonstrating that simplified models emphasizing mass distribution over granulochemical detail can still guide effective mine planning and processing without major losses in accuracy.