&Emsp Deep learning-based methods have attracted great attention in glacier extraction due to their advantages over traditional techniques. In this study, we verified the feasibility and effectiveness of LandsNet architecture in glacier extraction, and we applied the improved LandsNet (M-LandsNet) to extract the glacier contours in the headwaters of the Three Rivers. Two scenarios were compared using the band ratio method, U-Net, U-Net++, GlacierNet, SaU-Net, U-Net+cSE and LandsNet. The analysis of the two scenarios shows that M-LandsNet has the best performance and generalization ability among the 1986 methods.
| collect time | 1986/01/01 - 2021/12/31 |
|---|---|
| collect place | Sanjiangyuan headwaters region |
| data size | 7.6 MiB |
| data format | .shp |
| Coordinate system | |
| Projection | WGS84 |
&Emsp; Obtaining a reliable glacier inventory is the first step in training a deep learning network. In this study, SCGI is used as a real glacier profile and is considered as ground truth.
&Emsp;We further extracted the glacier outlines in the Three Rivers headwaters region in a total of 40 periods using M-LandsNet and manual adjustment.
&Emsp;Good quality of data
This work is licensed under a
Creative
Commons Attribution 4.0 International License.
| # | title | file size |
|---|---|---|
| 1 | GlacierOutlines.rar | 7.6 MiB |
| 2 | _ncdc_meta_.json | 4.5 KiB |
1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021
©Copyright 2005-. Northwest Institute of Eco-Environment and Resources, CAS.
Donggang West Road 320, Lanzhou, Gansu, China (730000)

