20th. Nansen Fellowship Ph.D. Candidate defended in St. Petersburg

Dr. Natalia Zakhvatkina has completed her studies on developing automatic routines for sea ice classification and monitoring using satellite radar sensors (SAR) in the Arctic Ocean. Her research has been completed under joint supervision from Arctic and Antarctic Research Institute (AARI) - Prof. Ivan E. Frolov and the Nansen Centers in St. Petersburg Dr. Vitaliy Yu. Alexandrov and Bergen - Prof. Ola M. Johannessen. She is the 20th Russian Nansen Fellowship Ph.D. candidate completing receiving a Doctoral degree in St. Petersburg.

A multilayer feed forward Neural Network (NN) algorithm is developed for the Arctic sea ice classification during the winter period. The algorithm can be applied to ENVISAT Advanced Synthetic Aperture Radar (ASAR) images using extracted backscatter coefficients and image texture features. Based on the visual interpretation of ASAR images, a neural network is trained for the classification of the first year (FY) level and rough ice and multiyear (MY) ice. The algorithm validation is done using Arctic and Antarctic Research Institution (AARI) ice charts and ice expert visual analysis. Preliminary neural network classification errors are 15% for level first year ice, 17% for deformed first year ice and 20% for multiyear ice (see Figure).

The backscatter coefficients for the major sea ice types at HH–polarization and 23° incidence angle, as well as angular dependencies of the backscatter for young, first-year and multiyear ice types are derived from calibrated ENVISAT ASAR Wide Swath Mode (WSM) images. A methodology is developed for the backscatter angular correction for the predetermined incidence angle for obtaining range independent contrast for the same ice types. The backscatter coefficient data sets for various Arctic winter sea ice types at HH–polarization and 23° incidence angle are derived from Envisat ASAR image analysis.

The Alternating Polarization images from ENVISAT ASAR have been analyzed over sea ice areas for various parts of the Arctic. The most promising results suggest using HH and VV polarization to discriminate ice and open water at high incidence angles (swath IS5 – IS7). Also classification of multiyear, first year and thin ice types can be improved by the use of polarization data. Some recommendations are given concerning the combination of co-and cross-polarization which is optimal for the ice type classification for better detection.

Figure caption: An example of the application of the developed algorithm for ENVISAT ASAR image classification. 18 January 2008, Central Arctic Ocean.
a) Raw ASAR WSM image (150-m resolution, the swath width is approximately 400 km).
b) Corrected, using elaborated methodology, image.
c) ASAR image classification using trained NN for three ice types derivation.
d) ASAR image classification – the result of consecutive application of the two trained NN for four sea ice types.
Colors of classified image (c and d): green – FY level ice, blue – FY deformed ice, red – MY ice, blue – calm open water/nilas.