UAV-Based Multispectral Assessment of Soil Water Content for Shallow Landslide Monitoring: A Machine Learning Approach

An increase in soil water content (SWC) from rainfall infiltration reduces the matric suction and shear strength; hence, rainfall is a primary trigger of shallow landslides. While accurate SWC monitoring is critical for predicting slope failure, traditional point-based sensors lack the spatial resolution required for effective field-scale assessment. This study aims to bridge this gap by integrating hyperspectral and multispectral imaging technologies with advanced machine learning (ML) models. Based on 114 in-situ soil samples collected from landslide-affected areas across South Korea, correlations between physical soil properties (e.g., void ratio, soil color) and hyperspectral data in the visible and near-infrared (Vis-NIR) regions were analyzed. Two ML algorithms, Random Forest (RF) and Multilayer Perceptron (MLP), were employed to develop predictive models for SWC. In this study, statistical evaluation indicated that the RF model demonstrated superior accuracy and robustness in handling high-dimensional spectral data compared to the MLP model. To validate the method's applicability for landslide monitoring, field tests were conducted in the mountainous region of Pyeongchang, South Korea, using a multispectral camera mounted on an unmanned aerial vehicle (UAV). The RF model successfully predicted the spatial distribution of SWC using spectral reflectance and geotechnical parameters. Although the model showed limitations in extrapolating beyond the training data range, it effectively captured critical variations in soil moisture relevant to slope stability. These results suggest that integrating UAV-based remote sensing with ML offers a promising, non-contact approach for high-resolution monitoring of shallow landslides, contributing to more proactive disaster prevention strategies.