Deep Learning-Based Remote Sensing Monitoring of Rock Glaciers—Preliminary Application in the Hunza River Basin

Abstract

Highlights: In this study, rock glacier identification was integrated with deep learning by constructing three distinct network architectures for semantic segmentation. The methodology was applied in the Hunza River Basin using Sentinel-2 MSI remote sensing imagery, achieving satisfactory rock glacier classification through iterative training. To improve recognition accuracy, geomorphological texture-based feature enhancement techniques were incorporated. Comparative analysis demonstrated that HRnet outperformed DeepLabV3+ in classification accuracy and surpassed U-Net in computational efficiency, making it more suitable for identifying the initiation and termination positions as well as boundaries of rock glaciers. This work presents a more effective approach for large-scale rock glacier identification and provides a reference for further applications of deep learning in cryospheric studies. What are the main findings? Three deep learning based semantic segmentation architectures (HRnet, DeepLabV3+, and U-Net) were constructed and evaluated for automatic rock glacier identification using Sentinel-2 MSI imagery in the Hunza River Basin. HRnet achieved the best overall performance, providing higher classification accuracy than DeepLabV3+ and greater computational efficiency than U-Net, particularly in delineating initiation zones, termination areas, and rock glacier boundaries. What are the main findings? The proposed approach offers an effective and scalable workflow for large-scale rock glacier mapping, reducing reliance on manual interpretation. This study provides a practical reference for applying deep learning to cryosphere monitoring and supports future automated geo-hazard assessment in high-mountain environments. Rock glaciers have been recognized as key indicators of geomorphic and climatic processes in high mountain environments. In this study, Sentinel-2 MSI imagery and topographic data were integrated to construct enhanced feature sets for rock glacier identification. Three state-of-the-art deep learning models (U-Net, DeepLabV3+, and HRnet) were employed to perform semantic segmentation for extracting rock glacier boundaries in the Hunza River Basin, located in the eastern Karakoram Mountains. The combination of spectral and terrain features significantly improved the differentiation of rock glaciers from surrounding landforms, establishing a robust basis for model training. A series of comparative experiments were conducted to evaluate the performance of each model. The HRnet model achieved the highest overall accuracy, exhibiting superior capabilities in high-resolution feature representations and generalization. Using the HRnet framework, a total of 597 rock glaciers were identified, covering an area of 183.59 km2. Spatial analysis revealed that these rock glaciers are concentrated between elevations of 4000 m and 6000 m, with maximum density near 5000 m, and a predominant south and southwest orientation. These spatial patterns reflect the combined influences of topography, thermal conditions, and snow accumulation on the formation and preservation of rock glaciers. The results confirm the effectiveness of deep learning-based semantic segmentation for large-scale rock glacier mapping. The proposed framework establishes a technical foundation for automated monitoring of alpine landforms and supports future assessments of rock glacier dynamics under climate variability.