Title:	Spectrally-based soil carbon models to support the National Forest Inventory of Rio de Janeiro, Brazil.
Authors:	VASQUES, G. M. LUZ, L. B. BALIEIRO, F. de C. MAGALHÃES, M. A. F. SILVEIRA FILHO, T. B. ANDRADE, M. T. de
Affiliation:	GUSTAVO DE MATTOS VASQUES, CNPS LEVI B. LUZ, UNIVERSIDADE FEDERAL FLUMINENSE FABIANO DE CARVALHO BALIEIRO, CNPS MONISE A․ F․ MAGALHÃES, SECRETARIA DE ESTADO DO AMBIENTE E SUSTENTABILIDADE DO RIO DE JANEIRO TELMO B. SILVEIRA FILHO, SECRETARIA DE ESTADO DO AMBIENTE E SUSTENTABILIDADE DO RIO DE JANEIRO MARCELO TEIXEIRA DE ANDRADE, CNPS.
Date Issued:	2026
Citation:	Soil Security, v. 23, 100236, Jun. 2026.
Description:	Methods for assessing soil carbon must be fast and accurate to support soil health and security evaluation, measurement, reporting and verification of carbon stocks. Visible-near infrared (Vis-NIR) spectroscopy minimizes the costs and time required for assessing soil carbon. Soil samples were obtained at 189 sites from the National Forest Inventory (NFI) of Rio de Janeiro state (∼​​43,782 km²), Brazil, between 2013 and 2016, at 0–20 and 30–50 cm, with a total of 373 recovered samples. The objective was to compare different preprocessing transformations of spectral curves, and calibration methods to predict soil carbon contents from soil Vis-NIR spectral curves in NFI samples from Rio de Janeiro state. Soil carbon contents were measured by dry combustion in a CHNS elemental analyzer, and soil spectral curves were obtained in the laboratory. Cubist combined with log(1/reflectance) preprocessing emerged as the optimal combination to predict soil carbon content (root mean square error of cross-validation of 5.1 g kg-1), whereas elastic net obtained good soil carbon content predictions consistently in both cross-validation and external validation. Partial least squares regression, random forest, support vector machines, and model ensemble produced poorer results. The results agree with previous studies comparing calibration methods for soil carbon content prediction, and stress the importance of preprocessing soil spectral curves, as well as testing different methods to produce robust results. Soil Vis-NIR spectroscopy can be used to assess soil carbon contents in Rio de Janeiro, supporting expedited and accurate soil carbon stock, and stock change assessments in future phases of the NFI.
Thesagro:	Carbono Solo
NAL Thesaurus:	Chemometrics
Keywords:	Soil spectral library Machine Learning Spectral modeling Biblioteca espectral de solo Aprendizado de máquina Quimiometria Modelagem espectral
DOI:	https://doi.org/10.1016/j.soisec.2026.100236
Type of Material:	Artigo de periódico
Access:	openAccess
Appears in Collections:	Artigo em periódico indexado (CNPS)