Passive microwave remote sensing of thin sea ice using principal component analysis

Abstract

Time sequences of surface based measurements of passive microwave emission from growing saline ice reported by Wensnahan et al. (1993) are used to explore the possibility of developing a satellite based sea ice concentration algorithm which solves for the presence of thinner ice. It is shown that two classes of thinner ice can be distinguished from mixtures of open water (OW), first‐year (FY) ice, and multiyear (MY) ice. The two classes do not necessarily correspond to specific World Meteorological Organization ice types; rather, newly formed ice represents a brief transition spectrum between OW and thin ice. Newly formed ice appears to be optically thick at 37 and 90 GHz and has a relatively dry surface. The thin ice spectrum occurs when the ice is greater than 4 cm thick and appears to result from the accumulation of brine at the surface of the ice. Thin ice has a relatively stable spectrum characterized by high brightness temperatures, a near‐zero spectral gradient at vertical polarization, and a large difference between vertical and horizontal polarizations. Supervised principal component analysis (PCA) was done of laboratory data using 10 channels of passive data: vertical and horizontal polarization at 6.7, 10, 19, 37, and 90 GHz. Analyses were also done on subsets of the laboratory data at 6.7 to 37 GHz as well as 19 to 90 GHz, representing the scanning multichannel microwave radiometer (SMMR) and special sensor microwave imager (SSM/I) satellite frequencies, respectively. Using all of the channels or the SMMR subset makes it possible to solve for mixtures of OW and FY, MY, newly formed and thin ice but with large errors. However, any four of these scene types can be distinguished with reasonable accuracy. The SSM/I frequencies allow determination of at most four of these scene types but with moderate errors. PCA was used in a case study of SSM/I data from the Bering Sea for April 2, 1988. Winds from the north formed thin ice areas which the NASA Team algorithm interprets as large amounts of OW and MY ice. With PCA, these same areas are interpreted as 20–30% OW near the lee shores but otherwise as consisting almost entirely of thin ice. We conclude that thin ice can be detected using satellite data. However, questions remain as to how the thin ice spectrum varies with environmental conditions, how it evolves to that of FY, and how this evolution affects the predicted concentrations of thin ice.

An algorithm to detect sea ice leads by using AMSR-E passive microwave imagery

“…However, the significant influence of the atmospheric variability restrains this approach. Wensnahan et al (1993) concluded that small areas of thin ice cannot be detected by passive microwave imagery due to their low resolution. The technique presented in this study uses the 89 GHz channel of the AMSR-E sensor with a considerable higher resolution than other microwave channels.…”

“…New ice, nilas, and pancake ice are denominated as thin ice according to the definition of the World Meteorological Organization (WMO, 1989). Wensnahan et al (1993) performed a principal component analysis of thin ice microwave emission spectra and concluded that thin ice can be distinguished from mixtures of open water, first-year ice, and multi-year ice by its spectral characteristics. Cavalieri (1994) presents a microwave technique to map thin ice.…”

An iterative approach to multisensor sea ice classification

“…The proposed classifier did not include a separate classification cluster for this ice type. A useful line of future research would include a study of the potential of segmenting frost flower-covered ice from RFY ice using the results of Wensnahan [7] and Ulander [28]. Fig.…”

“…Clearly, a majority of the data variance is contained in the top few principal component images, implying that lower PCA images can be neglected with minimal effect on the final classification. Wensnahan et al suggest keeping only PCA parameters whose variance is much larger than measurement uncertainty (converted into principal component space) [7]. Such a choice of eigenvectors allows information to be separated from noise.…”

“…Lower resolution techniques have also been proposed. Wensnahan et alproposed a classification method using passive radiometer data [7] to estimate the concentrations of first-year, multi-year, and thin ice in the Arctic. In [8], a classifier was developed that uses single channel 5.3 GHz ERS scatterometer data.…”