Refinamiento de dominios litológicos mediante modelos ocultos de Markov (HMM) y clasificación no paramétrica de datos geoquímicos con kriging ordinario

Abstract

Abstract: This study evaluated the refinement of lithological domains using Hidden Markov Models (HMMs) and its effect on the geostatistical estimation of geochemical data. A total of 4,000 samples and 37 elements were analyzed after quality control, centered log-ratio transformation, Z-score standardization, and principal component analysis. The 12 retained components explained 81.58% of the variance and were used for HDBSCAN clustering and HMM construction. The selected model generated six geochemical domains and increased sequential continuity from 64.36% to 88.00%, with a mean maximum posterior probability of 0.9635. Compared with the original lithological classification, the refined domains increased internal homogeneity for 36 of the 37 elements. Cross-validation of ordinary kriging reduced the RMSE for Ag, Cu, Mo, and Zn, with the greatest improvement observed for Cu (11.87%), whereas As and Pb showed no improvement. The findings indicate that HMM-derived domains complement rather than replace lithological interpretation, and that their practical value depends on the element being modeled.