Development and validation of regression models from NIR spectra to predict the composition of sugarcane, soybean meal, and cornmeal

• Autor(a) principal: Trópia, Nathália Veloso
• Data de Publicação: 2023
• Outros Autores: Silva, Flávia Adriane de Sales, Andrade, Dhones Rodrigues, Cidrini, Fernando Alerrandro Andrade, Ebani, Yuri Cesconetto, Matos, Éllem Maria de Almeida, Borges, Karen Melo, Roque, Jussara Valente, Zanetti, Diego, Valadares Filho, Sebastião de Campos
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Semina. Ciências Agrárias (Online)
• Texto Completo: https://ojs.uel.br/revistas/uel/index.php/semagrarias/article/view/47010
• topic: Chemometrics; Partial least squares regression; Spectroscopy.; Espectroscopia; Quimiometria; Regressão por mínimos quadrados parciais.

This study aimed to develop and assess regression models for predicting the chemical composition of sugarcane, soybean meal, and cornmeal using portable near-infrared (NIR) spectroscopy combined with chemometric techniques. A total of 95 sugarcane samples, 92 soybean meal samples, and 120 cornmeal samples were used. The samples were ground, and NIR spectra were obtained for each sample. Reference values were determined through conventional chemical analysis. Partial least squares regression and leave-one-out cross-validation were employed to construct the models. Models with the lowest root mean squared error in cross-validation were further validated externally. The goodness-of-fit of the models was evaluated by comparing the predicted values with those obtained through conventional laboratory methods. The constructed models properly estimated all constituents evaluated for sugarcane, soybean meal, and cornmeal (P ≥ 0.056). The models developed for predicting the contents of samples oven-dried at 55 °C (ADS) and 105 °C (ODS), total dry matter (DM), organic matter (OM), neutral detergent fiber (NDF), NDF corrected for ash and protein (NDFap), neutral detergent insoluble protein (NDIP), acid detergent fiber (ADF), crude protein (CP), non-fiber carbohydrates (NFC), and total digestible nutrients (TDN) in sugarcane; ODS, OM, NDF, ADF, indigestible NDF (iNDF), CP, TDN, and starch in soybean meal; and ODS and CP in cornmeal exhibited high accuracy and precision (R2 ≥ 0.50 and CCC ≥ 0.60). However, the models developed for predicting the levels of neutral detergent insoluble ash (NDIA) in sugarcane; ether extract (EE) and NDIA in soybean meal; and NDF, iNDF, NDIA, NFC, and EE in cornmeal demonstrated accuracy but lacked precision (R2 ≥ -0.04 and CCC ≥ 0.03). In conclusion, the portable NIR regression models provided accurate estimates and are therefore recommended for predicting the chemical composition of sugarcane, soybean meal, and cornmeal.