Deep learning-based method for the extraction of surface crevasse in mountain glaciers:a case study of Yanong Glacier in the southeastern Qinghai-Xizang Plateau
Understanding the dynamics of glaciers is of paramount importance in the context of global climate change,as glaciers serve as sensitive indicators of environmental shifts.They are key components of the Earth's freshwater reserves,influencing sea level,water resources,and the global climate system.The study of gla-ciers,therefore,extends beyond pure scientific inquiry;it has critical implications for predicting future water availability,assessing potential sea-level rise,and developing strategies for mitigating the impacts of climate change.Within this broader framework,glacier surface crevasses offer a window into the deeper processes at work within glaciers.These features are not merely physical anomalies but are indicative of the glacier's internal and external dynamics.They provide insights into the stress and strain glaciers undergo as they move and de-form,offering clues about the glacier's velocity,the distribution of stresses within the ice,and the interaction between the glacier and its underlying bedrock.Moreover,crevasses influence the energy balance of a glacier's surface by altering its albedo and affecting its meltwater dynamics,which in turn affects the glacier's overall mass balance and movement.Thus,the detailed study of crevasses contributes significantly to our understanding of glacier mechanics and their response to climatic variations.This study focuses on the Yanong Glacier,located in the Kangri Karpo Mountain of the southeastern Qinghai-Xizang Plateau,an area noted for its complex terrain and significant glaciological interest.The deployment of Unmanned Aerial Vehicles(UAVs),particularly the DJI M300 RTK UAV,has facilitated the acquisition of high-resolution orthophotos with a ground resolution of 0.03 m.This technological advancement enables the detailed mapping of glacier surface features,which is cru-cial for identifying and analyzing crevasse patterns and their distribution across the glacier's surface.To address the challenges associated with crevasse detection,the study employs a sophisticated deep learning framework,marking a significant advancement over traditional analytical methods.The introduction of the Convolutional Block Attention Module(CBAM)-UNet model represents a novel approach in the field of remote sensing and glaciology.This model enhances the feature extraction capabilities of the conventional U-Net architecture by in-corporating attention mechanisms that focus on relevant features for more accurate crevasse detection.The com-parative analysis conducted in this study demonstrates the superiority of the CBAM-UNet model over established deep learning models such as U-Net,DeeplabV3+,PSPNet,and HRNet.Achieving a precision rate of 90.74%in identifying ice crevasses,the CBAM-UNet model proves to be exceptionally effective in the extraction of vari-ous crevasse types,including Transverse,Splaying,En Échelon,Marginal Crevasses,and Rifts.These cre-vasse types,indicative of the glacier's dynamic behavior and influenced by the underlying topography,provide valuable information on the structural integrity and movement patterns of the glacier.The utilization of high-reso-lution UAV imagery in conjunction with advanced deep learning models offers a robust method for the detailed and accurate detection of glacier crevasses.This methodological approach not only enhances the precision of cre-vasse mapping but also facilitates the continuous monitoring of glacier transformations.Such advancements are instrumental in understanding the complex responses of glaciers to climate change,contributing essential knowl-edge to the fields of glaciology and climate science.Furthermore,this investigation sheds light on the potential applications of UAV technology and deep learning in glaciological research,highlighting the importance of inter-disciplinary approaches in addressing environmental challenges.The findings of this study underscore the utility of combining high-resolution remote sensing data with machine learning techniques to improve the accuracy and efficiency of glacier monitoring efforts.In conclusion,the research presented herein exemplifies the significant contributions of technological innovations to the study of glacier dynamics.By enhancing our ability to accurate-ly map and analyze glacier surface crevasses,this study offers valuable insights into glacier behavior,the impact of climate change on glacier systems,and the broader implications for water resources and sea-level rise.It is hoped that the methodologies and findings of this study will serve as a foundation for future research in glaciolo-gy,promoting further advancements in the understanding and preservation of these vital natural resources.
万方数据
