Espectroscopia de reflectância difusa no infravermelho próximo e aprendizado de máquina na determinação do teor de matéria orgânica de solos do Espírito Santo
Nenhuma Miniatura disponível
Data
2021-08-25
Autores
Souza, Thamiris Ferreira de
Título da Revista
ISSN da Revista
Título de Volume
Editor
Universidade Federal do Espírito Santo
Resumo
The research on the use of NIR spectroscopy to predict soil attributes, whether chemical or physical, has been receiving attention in recent years, because, in addition to presenting high potential as a soil analysis methodology, the technique has advantages when compared to methods used by routine laboratories, such as speed and practicality by dispensing with laborious analytical steps and the use of chemical reagents. Therefore, it contributes directly to the environment, as it does not use reagents often with high toxicity. The present study aimed to evaluate models with information from diffuse reflectance spectroscopy in the near infrared, for the prediction of organic matter contents in soil samples from the state of Espírito Santo. The following steps were followed: preparation of soil samples (Air-dried fine soil sieved in 2 mm); obtaining the organic matter contents and NIR spectra from the database; selection of sample groups and stratification with the K-means algorithm; application of pre-treatments; finally, adjustments and data evaluation were made to the PLS, Cubist, SVM, RF, GBM and GPR regression models. Among the results for the prediction of organic matter contents, R²=0.83 and R²=0.75 were highlighted, using the SG pre-processing technique associated with the PLS model and the IRLS technique with the Cubist model, respectively. These results come from the test set separated by the K-means algorithm. They were considered satisfactory. The stratification of the data group into subgroups proved to be effective in improving the predictive capacity of the built models. Cubist, linear GPR and polynomial SVM models also showed good performance for the prediction of soil organic matter associated with NIR spectroscopy, assuming a starting point for future studies.
Descrição
Palavras-chave
Quimiometria , aprendizado de máquina , matéria orgânica do solo , Spectroscopy NIR