Delineation of Glacier Margins with Satellite Images and Neural Networks
Climate change has led to the retreat of glaciers with the consequence that the area of the Icelandic glaciers has decreased by ca. 800 km2 since the year 2000. It is important to monitor these changes, e.g. by delineating the glacier margins. The aim of this study is to develop a method to delineat...
Main Author: | |
---|---|
Other Authors: | |
Format: | Master Thesis |
Language: | English |
Published: |
2023
|
Subjects: | |
Online Access: | http://hdl.handle.net/1946/44681 |
id |
ftskemman:oai:skemman.is:1946/44681 |
---|---|
record_format |
openpolar |
spelling |
ftskemman:oai:skemman.is:1946/44681 2023-06-18T03:41:38+02:00 Delineation of Glacier Margins with Satellite Images and Neural Networks Ákvörðun jökuljaðra með gervitunglamyndum og tauganetum Jón Ingimarsson 1996- Háskóli Íslands 2023-06 application/pdf http://hdl.handle.net/1946/44681 en eng http://hdl.handle.net/1946/44681 Tölvunarfræði Thesis Master's 2023 ftskemman 2023-06-07T22:53:22Z Climate change has led to the retreat of glaciers with the consequence that the area of the Icelandic glaciers has decreased by ca. 800 km2 since the year 2000. It is important to monitor these changes, e.g. by delineating the glacier margins. The aim of this study is to develop a method to delineate glacier margins automatically, and consequently be able to obtain updated glacier margins more frequently. Satellite images from Sentinel-2 from the 30th of September 2019 and manually digitized glacier margins based on these images from the Icelandic Meteorological Office were used. The study focuses on Langjökull and its small neighbouring glaciers, as well as on Mýrdalsjökull. The convolutional neural network architectures U-Net and DeepLabv3 were trained as models for automatic delineation of glacier margins. Two workflows were created, one that creates models and another that uses models for inference. Both workflows require preprocessing, while only one of them requires postprocessing to create shapefiles with glacier margins from the model. The model performance was different for the two glaciers. Langjökull appeared better suited for the methods developed in this study. The results of the automatic processing are accurate enough to be used as a first draft of a new glacier-margin dataset, which subsequently needs to be refined and corrected manually by a specialist. The model performance is best for glaciers with minimal or no debris cover but the model encounters problems for glacier with a thick debris layer on top of the glacier ice. Hlýnandi loftslag hefur ýtt undir hörfun jökla með þeim afleiðingum að flatarmál íslenskra jökla hefur minnkað um u.þ.b. 800 km2 frá árinu 2000. Mikilvægt er að fylgjast með þessum breytingum, t.d. með hnitun jökuljaðra. Markmið þessarar rannsóknar er að þróa sjálfvirka hnitun jökuljaðra á Íslandi sem gerir kleift að meta breytingar á útbreiðslu jökla oftar en ella væri mögulegt. Notaðar voru gervitunglamyndir frá Sentinel-2 gervihnöttum frá 30. september 2019 og ... Master Thesis Langjökull Mýrdalsjökull Skemman (Iceland) Langjökull ENVELOPE(-20.145,-20.145,64.654,64.654) Mýrdalsjökull ENVELOPE(-19.174,-19.174,63.643,63.643) New Glacier ENVELOPE(162.400,162.400,-77.033,-77.033) |
institution |
Open Polar |
collection |
Skemman (Iceland) |
op_collection_id |
ftskemman |
language |
English |
topic |
Tölvunarfræði |
spellingShingle |
Tölvunarfræði Jón Ingimarsson 1996- Delineation of Glacier Margins with Satellite Images and Neural Networks |
topic_facet |
Tölvunarfræði |
description |
Climate change has led to the retreat of glaciers with the consequence that the area of the Icelandic glaciers has decreased by ca. 800 km2 since the year 2000. It is important to monitor these changes, e.g. by delineating the glacier margins. The aim of this study is to develop a method to delineate glacier margins automatically, and consequently be able to obtain updated glacier margins more frequently. Satellite images from Sentinel-2 from the 30th of September 2019 and manually digitized glacier margins based on these images from the Icelandic Meteorological Office were used. The study focuses on Langjökull and its small neighbouring glaciers, as well as on Mýrdalsjökull. The convolutional neural network architectures U-Net and DeepLabv3 were trained as models for automatic delineation of glacier margins. Two workflows were created, one that creates models and another that uses models for inference. Both workflows require preprocessing, while only one of them requires postprocessing to create shapefiles with glacier margins from the model. The model performance was different for the two glaciers. Langjökull appeared better suited for the methods developed in this study. The results of the automatic processing are accurate enough to be used as a first draft of a new glacier-margin dataset, which subsequently needs to be refined and corrected manually by a specialist. The model performance is best for glaciers with minimal or no debris cover but the model encounters problems for glacier with a thick debris layer on top of the glacier ice. Hlýnandi loftslag hefur ýtt undir hörfun jökla með þeim afleiðingum að flatarmál íslenskra jökla hefur minnkað um u.þ.b. 800 km2 frá árinu 2000. Mikilvægt er að fylgjast með þessum breytingum, t.d. með hnitun jökuljaðra. Markmið þessarar rannsóknar er að þróa sjálfvirka hnitun jökuljaðra á Íslandi sem gerir kleift að meta breytingar á útbreiðslu jökla oftar en ella væri mögulegt. Notaðar voru gervitunglamyndir frá Sentinel-2 gervihnöttum frá 30. september 2019 og ... |
author2 |
Háskóli Íslands |
format |
Master Thesis |
author |
Jón Ingimarsson 1996- |
author_facet |
Jón Ingimarsson 1996- |
author_sort |
Jón Ingimarsson 1996- |
title |
Delineation of Glacier Margins with Satellite Images and Neural Networks |
title_short |
Delineation of Glacier Margins with Satellite Images and Neural Networks |
title_full |
Delineation of Glacier Margins with Satellite Images and Neural Networks |
title_fullStr |
Delineation of Glacier Margins with Satellite Images and Neural Networks |
title_full_unstemmed |
Delineation of Glacier Margins with Satellite Images and Neural Networks |
title_sort |
delineation of glacier margins with satellite images and neural networks |
publishDate |
2023 |
url |
http://hdl.handle.net/1946/44681 |
long_lat |
ENVELOPE(-20.145,-20.145,64.654,64.654) ENVELOPE(-19.174,-19.174,63.643,63.643) ENVELOPE(162.400,162.400,-77.033,-77.033) |
geographic |
Langjökull Mýrdalsjökull New Glacier |
geographic_facet |
Langjökull Mýrdalsjökull New Glacier |
genre |
Langjökull Mýrdalsjökull |
genre_facet |
Langjökull Mýrdalsjökull |
op_relation |
http://hdl.handle.net/1946/44681 |
_version_ |
1769007278713733120 |