Unknown
In:
XXVIII General Assembly of the International Union of Geodesy and Geophysics (IUGG)
Publication Date:
2023-08-16
Description:
Glacier calving is an essential process for the ablation of marine-terminating glaciers. The calving front position is needed to compute the glacier surface area and the total frontal ablation rate. With the increasing number of satellite observations, the calving process is observable with a higher temporal resolution. We avoid the labour-intensive annotation of the calving front in satellite images by training a neural network that delineates the calving fronts automatically. We use a deep learning framework called nnU-Net (no new U-Net) that adapts the widely used segmentation network U-Net to a given dataset. We evaluate the method on a benchmark dataset called CaFFe (CAlving Fronts and where to Find thEm). The dataset is conceived to compare different methods for calving front detection and glacier zone segmentation from Synthetic Aperture Radar images. In the end, we examine season and satellite's impact on the calving front detection quality.
Language:
English
Type:
info:eu-repo/semantics/conferenceObject
|
Location |
Call Number |
Expected |
Availability |
