Machine learning and hyperspectral imaging to predict soil water content: methodology and field validation

Accurate soil water content information is critical for effective land use and environmental management. However, many soil water content prediction methods lack sufficient real-scale validation and accuracy. This study aimed to combine hyperspectral and multispectral imaging technologies with advanced machine learning (ML) models to predict soil water content. 114 soil samples were directly collected from landslide-affected areas across South Korea. These samples were used to integrate hyperspectral data in the visible and near-infrared (NIR) regions to address the limitations of traditional soil water assessment methods. Two ML models, random forest (RF) and multilayer perceptron (MLP), renowned for their efficiency in processing complex high-dimensional data, were used to develop robust predictive models, which had their performances evaluated. Statistical analyses showed that the RF model was more accurate than the MLP model. Field validation testing of the RF model was conducted in the mountainous region of Pyeongchang. A multispectral camera was mounted on an unmanned aerial vehicle to enhance the practical applicability of the model. The RF model reliably predicted soil water content by effectively handling spectral data input, including data on the soil color, void ratio, and reflectance area in NIR. However, the RF model was not well-suited for extrapolation beyond the training data range. The proposed method will support agricultural management and disaster prevention and contribute to better land management practices. The study results demonstrate the potential for integrating hyperspectral imaging with ML techniques to enhance the precision and reliability of environmental monitoring tools, offering valuable insights for sustainable land management and ecological conservation.