On SAR-based statistical ice thickness estimation in the baltic sea

Abstract

In the Baltic Sea the annual maximum ice extent varies strongly, from the minimum (12 %) to an almost total ice cover over the Baltic Sea, the average being 40 % during the last 30 years. The occurrence of such large annual ice cover has guided the SAR based sea ice mapping work done at the Finnish Institute of Marine Research (FIMR). During the winters 2003-2005 ice thickness measurement campaigns with a simultaneous acquisition of satellite-borne SAR data have been performed. Using these data sets we can address the problem how well one can estimate ice thickness or some other quantity related to it using solely SAR data. In this paper some estimation results for a 1-dimensional data set are presented. In our other paper [1] an approach to the estimation in the two dimensional setting, a much harder problem, is described. Here we will examine the applicability of two different quantities describing the total sea ice thickness. The quantities are: total ice thickness (T ) and equivalent deformed ice thickness (Tdef ). The proposed methods use either a classification or a regression approach. The surface scattering dominates the backscattering at Cband [2]. The link between the ice thickness and the backscattering strength is surface roughness. Typically the surface of level ice becomes more rough when aging due to weathering. Hence, the radar response is, on average, stronger from older (and thicker) level ice than from newer (and thinner) level ice. A major exception for this correspondence is the fast ice area where the thickest level ice fields occur. For many reasons, e.g. for snow ice, the surface roughness and ice thickness in the fast ice field are so weakly correlated that there does not exist any functional relationship between them. Another major exception is the marginal ice zone. In the sequel we restrict our analysis only to the estimation of ice thickness in drift ice areas, excluding fast ice and marginal ice zones, see [1]. The most drastic changes in the ice thickness are associated with the deformation of ice fields, i.e. the formation of ice ridges or hummocks. Both these deformation types significantly modify the ice surface characteristics. Hence, these deformed areas can be detected by SAR, the detection accuracy being dependent on the resolution of the sensor.