Geo-Environmental Risk Assessment of Sand Dunes Encroachment Hazards in Arid Lands Using Machine Learning Techniques

Machine Learning Techniques (MLTs) and accurate geographic mapping are crucial for managing natural hazards, especially when monitoring the movement of sand dunes. This study presents the integration of MLTs with geographic information systems (GIS) and “R” software to monitor sand dune movement in Najran City, Saudi Arabia (KSA). Utilizing Linear Support Vector Machine (SVM), Random Forest (RF), and Artificial Neural Networks (ANN) with nine dune-related variables, this study introduces a new Drifting Sand Index (DSI) for effectively identifying and mapping dune accumulations. The DSI incorporates multispectral sensors data and demonstrates a robust capability for monitoring sand dune dynamics. Field surveys and spatial data analysis were used to identify about 100 dune locations, which were then divided into training (70%) and validation (30%) sets at random. These models produced a thorough dune encroachment risk map that divided areas into five hazard zones: very low, low, medium, high, and very high risk. The results show an average sand dune movement of 0.8 m/year towards the southeast. Performance evaluation utilizing the Area Under Curve-Receiver Operating Characteristic (AUC-ROC) approach revealed AUC values of 96.2% for SVM, 94.2% for RF, and 93% for ANN, indicating RF (AUC = 96.2%) as the most effective MLTs. This crucial information provides valuable insights for sustainable development and environmental protection, enabling decision-makers to prioritize regions for mitigation techniques against sand dune encroachment.