Glaciernet Variant for Large Scale Glacier Mapping

Date of Award


Degree Name

Ph.D. in Electrical Engineering


Department of Electrical and Computer Engineering


Vijayan Asari


Climate change impact is profoundly visible in recent decades including its effect on the health of the mountain glaciers. Accelerated glacier recession patterns observed globally are leading to consequences like sea level rise, water security and glacier-related hazards. Therefore, it is important to monitor and understand these glacier changes. Detection of accurate glacier boundary, which is the basic input of many glacier analysis, remains a challenge even after many years of research on conventional remote sensing methods or machine learning methods. A deep learning based approach named as GlacierNet has been developed to exploit the convolutional neural network (CNN) segmentation model to accurately outline the debris-covered glacier (DCG) ablation zones in regional scope. To improve the approach, the performance of GlacierNet's CNN is compared with several advanced CNN segmentation models, including Mobile-UNet, Res-UNet, FCDenseNet, R2UNet, and DeepLabV3+, to identify the most salient features that could contribute to the DCG segmentation accuracy. Based on the evaluation, we developed the GlacierNet2 that is an enhanced version of our GlacierNet, that incorporates deep learning, image processing, and remote sensing technologies and hydrology science. It is observed that the GlacierNet2 ameliorates the estimation of the DCG ablation region also called the ablation zone and reaches the high level of intersections over union (IOU) score of 0.8839. The important component of any glacier mapping is to include both accumulation and ablation zone. Consequently, the newly added capacity of the enhanced approach is to map the snow-covered accumulation zone (SCAZ). The experimental evaluations demonstrate that the proposed model can provide complete glacier (both accumulation and ablation zone) outlines at regional scales with an overall IOU score of 0.8619. Also, we design a large-scale mapping strategy to progressively enhance the network familiarity to varied types of glaciers via systematically repeating the training process. This strategy allows the network to delineate a large number of glaciers while only requiring a small proportion of initial training data. Thus, resulting in a significant drop in labor and expert intervention, which are required for selecting and labeling the training data. Our results show a successful and accurate generation of glacier boundaries with an intersection over union (IOU) score of 0.8115 in the Karakoram and parts of western Himalaya and an IOU of 0.7525 in the Nepal Himalaya. Our work outlines how future efforts of large and global scale mapping can be developed to monitor and analyze glacier dynamics.


Electrical Engineering, Computer Science, Geographic Information Science, Remote Sensing, Deep learning, Image segmentation, Convolutional neural network, Glacier mapping, Remote sensing, Machine learning, GlacierNet

Rights Statement

Copyright © 2022, author