Título:	A non-destructive approach using spectral fingerprinting and chemometrics in the authentication and quality prediction of fallen tree wood.
Autoria:	NASCIMENTO, C. S. do ANDRADE, J. C. de SILVA, C. E. da KARTNALLER, V. CAMPOS, M. A. A. ARAÚJO, R. D. de FIGUEIREDO, A. da S. SOARES, J. C. R. OLIVEIRA, M. C. R. de HIGUCHI, N.
Afiliação:	CRISTIANO S. DO NASCIMENTO, INSTITUTO NACIONAL DE PESQUISAS DA AMAZÔNIA; JELMIR C. DE ANDRADE, UNIVERSIDADE FEDERAL DO RIO DE JANEIRO; CLAUDIA EUGENIO DA SILVA, INSTITUTO NACIONAL DE PESQUISAS DA AMAZÔNIA; VINICIUS KARTNALLER, UNIVERSIDADE FEDERAL DO RIO DE JANEIRO; MOACIR ALBERTO A. CAMPOS, INSTITUTO NACIONAL DE PESQUISA DA AMAZÔNIA; ROBERTO D. DE ARAÚJO, INSTITUTO NACIONAL DE PESQUISAS DA AMAZÔNIA; ADRYA DA S. FIGUEIREDO, INSTITUTO NACIONAL DE PESQUISAS DA AMAZÔNIA; JOSE CARLOS RODRIGUES SOARES, CPAA; MARCIA C. RAMOS DE OLIVEIRA, FEDERAL INSTITUTE OF EDUCATION, SCIENCE AND TECHNOLOGY OF AMAZONAS; NIRO HIGUCHI, INSTITUTO NACIONAL DE PESQUISAS DA AMAZÔNIA.
Ano de publicação:	2026
Referência:	Microchemical Journal, v. 225, art. 117971, June 2026.
Conteúdo:	The Amazon rainforest harbors a stock of naturally fallen trees whose rational use is limited mainly by the difficulty of accurately identifying their taxa and, consequently, assigning technological value to this resource. This study proposes a rapid, non-destructive strategy that authenticates the origin of these woods and predicts their quality by integrating Fourier Transform Near-Infrared spectroscopy with chemometrics. Exploratory analysis applied to spectra from ten Central Amazon taxa revealed clear spectral patterns for some species, especially in bands at 5284–5426 and 7123–7359 cm−1, associated with Csingle bondO and Csingle bondH vibrations of components such as lignin and hemicellulose. Using these fingerprints, a one-class classification approach implemented in MATLAB authenticated the woods with precision of 99.1–100%. Once identity was secured, the NIR spectra were used in PLS regression models developed in TQ Analyst software to predict properties chemical, physical and mechanical. Although the classification and regression steps were performed in different software environments due to the specific requirements of each algorithm, they were applied sequentially to the same spectral dataset, forming an integrated workflow. The models showed strong predictive ability, with coefficients of determination for prediction generally above 0.88, low prediction errors, and ratio of performance to deviation values mostly above 3, indicating suitability for screening and quantitative estimation. However, due to severe logistical constraints of sampling naturally fallen trees in the Amazon, these models were built with a limited test set and rely solely on internal validation; thus, these high-performance metrics should be interpreted cautiously as an exploratory proof-of-concept. While this study successfully demonstrates the feasibility of the method, rigorous external validation with larger, independent datasets is required before full implementation.
NAL Thesaurus:	Forest management
Palavras-chave:	Nondestructive tool Partial least squares regression Amazonian woods Technological properties Manejo florestal Wood authentication
Digital Object Identifier:	https://doi.org/10.1016/j.microc.2026.117971
Tipo do material:	Artigo de periódico
Acesso:	openAccess
Aparece nas coleções:	Artigo em periódico indexado (CPAA)