Silica (Si) in plants has become a subject of increasing interest for various communities, e.g., those involved in global change biogeochemistry, agronomy and biotechnology, and many techniques have been applied to quantitatively determine the Si content of plants. However, there are few applicable techniques for the in-situ assessment of Si contents because many methods are destructive and expensive. As a recent alternative, spectrophotometric analyses using hyperspectral data are both non-destructive and comparatively cheap, and machine learning algorithms have been used to enhance hyperspectral remote sensing to evaluate biochemical properties. The objectives of this study were to examine the potential of hyperspectral remote sensing approaches to estimate the silica content of Zizania latifolia using conventional machine learning algorithms including random forests, extreme gradient boosting and Cubist. The results indicate that Cubist was the best algorithm, achieving a ratio of performance to deviation of 2.02 and a root mean square error of 42.52 µg cm− 2.
Nofrizal et al. (Fri,) studied this question.