Abstract
The development of rapid techniques, such as hyperspectral spectrophotometry, for investigating arsenic (As) soil contamination could be of great value with respect to conventional methods. This study was conducted to detect As concentrations in artificially polluted soils (from 25 to 1045 mg kg−1) through hyperspectral visible–near infrared spectrophotometry and to compare two multivariate statistical regression analyses: partial least squares and support vector machines. The correlation coefficient r is greater in the partial least squares in both model (0.93%) and test (0.87%) with respect to support vector machines (0.88% for the model and 0.82% for the test). The most important model variables extracted from the variable importance in projection scores resulted the absorption peaks at 580, 660, 715, and 780 nm. Bands in the visible spectra are not directly associated to As, but the metalloid can interact with the main spectrally active components of soil permitting to multivariate statistical models to screen As concentrations.