偏最小二乘回归
化学
化学计量学
土壤水分
土工试验
分析化学(期刊)
漫反射红外傅里叶变换
线性回归
环境化学
矿物学
土壤科学
数学
环境科学
色谱法
统计
光催化
生物化学
催化作用
作者
Sławomir Krzebietke,Michal Daszykowski,H. Czarnik-Matusewicz,I. Stanimirova,L. Pieszczek,S. Sienkiewicz,J. Wierzbowska
出处
期刊:Talanta
[Elsevier]
日期:2022-07-01
卷期号:251: 123749-123749
标识
DOI:10.1016/j.talanta.2022.123749
摘要
This study illustrates the successful application of near-infrared reflectance spectroscopy extended with chemometric modeling to profile Cd, Cu, Pb, Ni, Cr, Zn, Mn, and Fe in cultivated and fertilized Haplic Luvisol soils. The partial least-squares regression (PLSR) models were built to predict the elements present in the soil samples at very low contents. A total of 234 soil samples were investigated, and their reflectance spectra were recorded in the spectral range of 1100–2500 nm. The optimal spectral preprocessing was selected among 56 different scenarios considering the root mean squared error of prediction (RMSEP). The partial robust M-regression method (PRM) was used to handle the outlying samples. The most promising models were obtained for estimating the amount of Cu (using PRM) and Pb (using the classic PLS), leading to RMSEP expressed as a percentage of the response range, equal to 9.63% and 11.5%, respectively. The respective coefficients of determination for validation samples were equal to 0.86 and 0.58, respectively. Assuming similar variability of model residuals for the model and test set samples, coefficients of determination for validation samples were 0.94 and 0.89, respectively. Moreover, the favorable PLS models were also built for Zn, Mn, and Fe with coefficients of determinations equal to 0.87, 0.87, and 0.79. • Monitoring the concentrations of eight elements in cultivated Haplic Luvisol soils. • NIR and determination of elements in soils at low concentrations. • Robust PLS modeling in the presence of outlying samples.
科研通智能强力驱动
Strongly Powered by AbleSci AI