MERCATOR OCEAN JOURNA: 
SEPTEMBER 2021 
These ongoing requirements will underpin the growing 
ınterest for these products and the need to expand the 
area covered. The availlability of more Sentinel-1 imagery 
means new automatic methods to derive ice edge position 
information need to be developed, supported by current 
widespread adoption of Artificial Intelligence (Al) and 
Machine Learning (ML) methods. 
1.5 Arctic Sentinel-1 ice type 
The Nansen Environmental and Remote Sensing Center 
NERSC) developed a new high-resolution automatic sea 
ice type (stage of development) product from Sentinel-1 
SAR data using a convolutional neural network (CNN) 
(Boulze et al., 2020). It provides classification of SAR data 
into four different types: open water, young ice, first-year 
ice, multi-year ice. The CNN was trained on manual ice 
charts prepared by the National Ice Center NOAA. CNN also 
provides probability of classification, which can be used as 
a measure of uncertainty. The daily NRT production was 
included in the MET-Norway operating system and made 
available to the Copernicus Marine Service as a dailv 
mosaic product covering the European part of the Arctic. 
% 
ce hype 2021-01-13 
* Open water 
Young ce 
Firspyear Ice 
Old ce 
{29 
255 
lat-lon ünes 
pr]! |[® 
„_ Example of Sentinel-1 sea ice tvpe product from 13. Januarv 2027 
1.6 Arctic multisensor-based sea ice 
concentration product 
MET Norway has developed an automatic multisensor sea 
ice concentration (SIC) product for the Arctic which is (by 
the end of Copernicus-1) provided as a daily SIC product. 
The new product has estimated SIC based on a combination 
of JAXA'’s AMSR2 microwave radiometer and C-band dual- 
polarized (HH/HV polarization combination) SAR data from 
Sentinel-1 
zach dataset is first classified as a SIC product before they 
are merged into one product using a variational merging 
process. The SAR data are first classified into ice/water at 
a 40m resolution. After the separation, the concentration Is 
calculated from the surface area of ice within a 1x1 km area. 
The AMSR2 SIC processing is based on the OSISAF/ ESA 
SCI+ Hybrid Dynamic (OSHD) algorithm which provides a 
spatial resolution of -5 km. Before merging, the AMSR2 SIC 
product is subsampled into 1x1 km resolution to match the 
SAR SIC. An example of the final product is shown below.