Stable crop yields require an appropriate supply of essential soil nutrients such as nitrogen (N), phosphorus (P), and potassium (K) based on the accurate diagnosis of soil nutrient status. Traditional laboratory analysis of soil nutrients is often complicated and time-consuming and does not provide real-time nutrient status. Visible-near-infrared (Vis-NIR) spectroscopy has emerged as a non-destructive and rapid method for estimating soil nutrient levels. Vis-NIR spectra reflect sample characteristics as the peak intensities; however, they are often affected by various artifacts and complex variables. Since Vis-NIR spectroscopy does not directly measure nutrient levels in soil, improving estimation accuracy is essential. For spectral preprocessing, the most important aspect is to develop an appropriate preprocessing strategy based on the characteristics of the data and identify artifacts such as noise, baseline drift, and scatter in the spectral data. Machine learning-based modeling techniques such as partial least-squares regression (PLSR) and support vector machine regression (SVMR) enhance estimation accuracy by capturing complex patterns of spectral data. Therefore, this review focuses on the use of Vis-NIR spectroscopy for evaluating soil properties including soil water content, organic carbon (C), and nutrients and explores its potential for real-time field application through spectral preprocessing and machine learning algorithms. Vis-NIR spectroscopy combined with machine learning is expected to enable more efficient and site-specific nutrient management, thereby contributing to sustainable agricultural practices.
Shin et al. (Thu,) studied this question.