GlacierNet2: A hybrid Multi-Model learning architecture for alpine glacier mapping

• DOI: 10.1016/j.jag.2022.102921
• Publication Date: 2022-07-20T16:40:03Z
• Authors: Zhiyuan Xie; Umesh K. Haritashya; Vijayan K. Asari; Michael P. Bishop; Jeffrey S. Kargel; Theus H. Aspiras
• Keywords: FOS: Computer and information sciences; Physical geography; Computer Science - Machine Learning; Electrical and Electronics; Alpine glacier; 0211 other engineering and technologies; 02 engineering and technology; 01 natural sciences; Machine Learning (cs.LG); Glacier mapping; GlacierNet; Systems and Communications; Other Electrical and Computer Engineering; FOS: Electrical engineering, electronic engineering, information engineering; GE1-350; Computer Engineering; 0105 earth and related environmental sciences; Image and Video Processing (eess.IV); Optics; Electrical and Computer Engineering; Remote sensing; Electrical Engineering and Systems Science - Image and Video Processing; 15. Life on land; GB3-5030; Environmental sciences; Debris-covered; 13. Climate action; Snow-covered; Electromagnetics and Photonics

In recent decades, climate change has significantly affected glacier dynamics, resulting in mass loss and an increased risk of glacier-related hazards including supraglacial proglacial lake development, as well catastrophic outburst flooding. Rapidly changing conditions dictate the need for continuous detailed observations analysis climate-glacier dynamics. Thematic quantitative information regarding geometry is fundamental understanding forcing sensitivity glaciers to change, however, accurately mapping debris-cover (DCGs) notoriously difficult based upon use spectral conventional machine-learning techniques. The objective this research improve earlier proposed deep-learning-based approach, GlacierNet, which was developed exploit a convolutional neural-network segmentation model outline regional DCG ablation zones. Specifically, we enhanced GlacierNet2 architecture thatincorporates multiple models, automatic post-processing, basin-level hydrological flow techniques DCGs such that it includes both accumulation Experimental evaluations demonstrate improves estimation zone allows high level intersection over union (IOU: 0.8839) score. provides complete (both zone) outlines at scales, with overall IOU score 0.8619. This crucial first step automating can be used accurate modeling or mass-balance analysis.