Predicting dune migration risks under climate change context: A hybrid approach combining machine learning, deep learning, and remote sensing indices

Given the impacts of climate change on increasing aridity, dune migration, and associated risks to adjacent areas and air quality, assessing these hazards is critical for effective land management. This study aims to utilize machine learning and deep learning algorithms to enhance image quality and delineate sand dune extents, identify optimal scales for extracting dune morphometric features, predict dune migration, and forecast climatic parameters and their relationships with morphometric characteristics. Results demonstrate that the deep iterative fusion network model effectively improves image quality for extracting dunes and their morphometric features with high accuracy. Furthermore, integrating morphometric and spectral features into a novel Land-Use Land-Form (LULF) map enables precise identification of landforms and objects in desert environments, including sand dune extents, with high accuracy. The findings also indicate that variations in spectral reflectance, particularly albedo and infrared bands, influence not only dune height detection but also dune migration speed. Additionally, the Markov model results suggest that increased albedo and infrared reflectance in the coming years will heighten the risk of dune migration in surrounding areas. Finally, the autoregressive integrated moving average model predicts future wind speeds ranging from 8.3 to 83.3 km/h, moving from southeast to northwest, reflecting intensified dune migration and increased risks to adjacent regions.