Using Deep Learning in Operational Data Products - Lessons Learned from the IceLines Dataset on Antarctic Ice Shelf Front Change
Antarctica’s ice shelves are the floating extensions of the ice sheet. The discharge of the Antarctic ice sheet increases if ice shelf areas with strong buttressing forces are lost. This has direct implications on Antarctica’s contribution to global sea level rise. Therefore, it is important to have...
Main Authors: | , , , |
---|---|
Format: | Conference Object |
Language: | English |
Published: |
2023
|
Subjects: | |
Online Access: | https://elib.dlr.de/196714/ https://elib.dlr.de/196714/1/2023-06-26_IceLines-IUGG_CB.pdf |
id |
ftdlr:oai:elib.dlr.de:196714 |
---|---|
record_format |
openpolar |
spelling |
ftdlr:oai:elib.dlr.de:196714 2024-05-19T07:30:12+00:00 Using Deep Learning in Operational Data Products - Lessons Learned from the IceLines Dataset on Antarctic Ice Shelf Front Change Baumhoer, Celia Dietz, Andreas Haug, Jan-Karl Künzer, Claudia 2023 application/pdf https://elib.dlr.de/196714/ https://elib.dlr.de/196714/1/2023-06-26_IceLines-IUGG_CB.pdf en eng https://elib.dlr.de/196714/1/2023-06-26_IceLines-IUGG_CB.pdf Baumhoer, Celia und Dietz, Andreas und Haug, Jan-Karl und Künzer, Claudia (2023) Using Deep Learning in Operational Data Products - Lessons Learned from the IceLines Dataset on Antarctic Ice Shelf Front Change. IUGG 2023, 2023-07-11 - 2023-07-20, Berlin, Germany. cc_by Dynamik der Landoberfläche Informationstechnik Konferenzbeitrag NonPeerReviewed 2023 ftdlr 2024-04-25T01:07:17Z Antarctica’s ice shelves are the floating extensions of the ice sheet. The discharge of the Antarctic ice sheet increases if ice shelf areas with strong buttressing forces are lost. This has direct implications on Antarctica’s contribution to global sea level rise. Therefore, it is important to have an operational product constantly providing data on ice shelf front position to locate and track changes in ice shelf area. Here, we present the workflow of the IceLines dataset showcasing a processing pipeline from acquired satellite data to a deep learning (DL) derived data product. The workflow includes the following steps: (1) triggering data download (2) pre-processing of Sentinel-1 SAR data with Docker on a high-performance cluster (3) training a convolutional neural network (CNN) for different input data formats (4) inference for ice shelf front detection (5) post-processing of the CNN output (6) sanity check of front positions based on the existing time series (7) automated data release via a web map service for data download and visualization. This contribution summarizes the lessons learned from implementing an DL-based operational data product including the challenges of big data processing, creating spatial and temporal transferable CNNs for image classification, detecting erroneous DL predictions and making geospatial datasets available to the public. Conference Object Antarc* Antarctic Ice Sheet Ice Shelf Ice Shelves German Aerospace Center: elib - DLR electronic library |
institution |
Open Polar |
collection |
German Aerospace Center: elib - DLR electronic library |
op_collection_id |
ftdlr |
language |
English |
topic |
Dynamik der Landoberfläche Informationstechnik |
spellingShingle |
Dynamik der Landoberfläche Informationstechnik Baumhoer, Celia Dietz, Andreas Haug, Jan-Karl Künzer, Claudia Using Deep Learning in Operational Data Products - Lessons Learned from the IceLines Dataset on Antarctic Ice Shelf Front Change |
topic_facet |
Dynamik der Landoberfläche Informationstechnik |
description |
Antarctica’s ice shelves are the floating extensions of the ice sheet. The discharge of the Antarctic ice sheet increases if ice shelf areas with strong buttressing forces are lost. This has direct implications on Antarctica’s contribution to global sea level rise. Therefore, it is important to have an operational product constantly providing data on ice shelf front position to locate and track changes in ice shelf area. Here, we present the workflow of the IceLines dataset showcasing a processing pipeline from acquired satellite data to a deep learning (DL) derived data product. The workflow includes the following steps: (1) triggering data download (2) pre-processing of Sentinel-1 SAR data with Docker on a high-performance cluster (3) training a convolutional neural network (CNN) for different input data formats (4) inference for ice shelf front detection (5) post-processing of the CNN output (6) sanity check of front positions based on the existing time series (7) automated data release via a web map service for data download and visualization. This contribution summarizes the lessons learned from implementing an DL-based operational data product including the challenges of big data processing, creating spatial and temporal transferable CNNs for image classification, detecting erroneous DL predictions and making geospatial datasets available to the public. |
format |
Conference Object |
author |
Baumhoer, Celia Dietz, Andreas Haug, Jan-Karl Künzer, Claudia |
author_facet |
Baumhoer, Celia Dietz, Andreas Haug, Jan-Karl Künzer, Claudia |
author_sort |
Baumhoer, Celia |
title |
Using Deep Learning in Operational Data Products - Lessons Learned from the IceLines Dataset on Antarctic Ice Shelf Front Change |
title_short |
Using Deep Learning in Operational Data Products - Lessons Learned from the IceLines Dataset on Antarctic Ice Shelf Front Change |
title_full |
Using Deep Learning in Operational Data Products - Lessons Learned from the IceLines Dataset on Antarctic Ice Shelf Front Change |
title_fullStr |
Using Deep Learning in Operational Data Products - Lessons Learned from the IceLines Dataset on Antarctic Ice Shelf Front Change |
title_full_unstemmed |
Using Deep Learning in Operational Data Products - Lessons Learned from the IceLines Dataset on Antarctic Ice Shelf Front Change |
title_sort |
using deep learning in operational data products - lessons learned from the icelines dataset on antarctic ice shelf front change |
publishDate |
2023 |
url |
https://elib.dlr.de/196714/ https://elib.dlr.de/196714/1/2023-06-26_IceLines-IUGG_CB.pdf |
genre |
Antarc* Antarctic Ice Sheet Ice Shelf Ice Shelves |
genre_facet |
Antarc* Antarctic Ice Sheet Ice Shelf Ice Shelves |
op_relation |
https://elib.dlr.de/196714/1/2023-06-26_IceLines-IUGG_CB.pdf Baumhoer, Celia und Dietz, Andreas und Haug, Jan-Karl und Künzer, Claudia (2023) Using Deep Learning in Operational Data Products - Lessons Learned from the IceLines Dataset on Antarctic Ice Shelf Front Change. IUGG 2023, 2023-07-11 - 2023-07-20, Berlin, Germany. |
op_rights |
cc_by |
_version_ |
1799484417315438592 |