Szczegóły publikacji
Opis bibliograficzny
Investigation of machine learning algorithms to determine glaciers displacements / Magdalena ŁUCKA // Remote Sensing Applications: Society and Environment [Dokument elektroniczny]. — Czasopismo elektroniczne ; ISSN 2352-9385. — 2025 — vol. 37 art. no. 101476, s. 1-16. — Wymagania systemowe: Adobe Reader. — Bibliogr. s. 14-16, Abstr. — Publikacja dostępna online od: 2025-01-31
Autor
Słowa kluczowe
Dane bibliometryczne
| ID BaDAP | 158981 |
|---|---|
| Data dodania do BaDAP | 2025-04-16 |
| Tekst źródłowy | URL |
| DOI | 10.1016/j.rsase.2025.101476 |
| Rok publikacji | 2025 |
| Typ publikacji | artykuł w czasopiśmie |
| Otwarty dostęp |  |
| Creative Commons | |
| Czasopismo/seria | Remote Sensing Applications: Society and Environment |
Abstract
Ice velocity changes in polar regions reflects the response of glacial areas to climate change progression. Its monitoring improves our understanding of the complexity of processes that affect this part of the world. Existing methods have some limitations like a need for extensive parametrization or adjustment of specific software to only one type of input data. Despite the rapid development of artificial intelligence, the fusion of satellite radar images and machine learning methods is still not thoroughly investigated topic. This study evaluates potential of using machine learning models and Sentinel-1 images for purposes of ice velocity determination. This approach is examined with two testing datasets and one real-world case study. The testing datasets contain original image and its manually shifted version to evaluate effectiveness and accuracy of proposed methodology. The real-world case study is Daugaard-Jensen glacier in eastern Greenland and its results are compared with the traditional remote sensing technique such as an offset-tracking algorithm. This research tested various data augmentation techniques and neural network architectures to find optimal solution that could serve as an alternative approach to deliver horizontal displacement information. The testing datasets showed that obtaining 1-pixel accuracy is possible with the proposed approach. Moreover, in real world conditions, the proposed methodology allowed for obtaining satisfactory results, showing also current limitations of presented approach related mainly to lack of distinguishing features in the main glacier flow.
