Atrous spatial pyramid pooling enhanced CNN for SAR based glacier facies segmentation

• The study proposes an automated glacier facies classification framework using SAR data and a CNN with ASPP for enhanced feature extraction. • The model achieves perfect precision (1.00) and high recall values (0.95-0.97) for key glacier zones like Wet Snow and Bare Ice Zones. • Debris-Covered Ice Zone and Frozen Percolation Zone show balanced precision and recall, with high F1-scores. • SAR and DEM data integration boosts classification accuracy, offering a scalable tool for glacier dynamics monitoring and climate research. Glacier facies provide important information about melting and refreezing patterns of glaciers. Advancement in glacier facies monitoring is pivotal for understanding the glacier behaviour under changing climate. The present study proposes an automated frame work for the segmentation of glacier facies Wet Snow Zone (WSZ), Bare Ice Zone (BIZ), Debris-Covered Ice Zone (DCIZ), and Frozen Percolation Zone (FPZ). The study evaluates the Atrous Spatial Pyramid Pooling (ASPP) enhanced Convolutional Neural Network (CNN) architecture for synthetic aperture radar (SAR) based facies segmentation. The model has performed well and has achieved the precision 1.00 for WSZ and recall value of 0.95 and 0.97 for WSZ and BIZ respectively. The model has attained an average recall of 0.99 and accuracy of 0.96 across all classes, indicating its sensitivity in class identification. The incorporation of the ASPP module has markedly enhanced the model's performance by extracting characteristics at different scales and maintaining both local and global context of the data. This study develops a scalable and effective solution for automating glacier facies segmentation, enhancing overall understanding and monitoring of glacial features.