Sea Ice Detection in the Sea of Okhotsk Using PALSAR and MODIS Data

Transcription

1 1516 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 6, NO. 3, JUNE 2013 Sea Ice Detection in the Sea of Okhotsk Using PALSAR and MODIS Data Hiroyuki Wakabayashi, Member, IEEE, Yuta Mori, and Kazuki Nakamura, Member, IEEE Abstract In this paper, we propose a new method for improving detection of sea ice through use of PALSAR (Phased-Array type L-band SAR) polarimetric data. Unlike traditional methods that are based on backscattering coefficient, our proposed method utilizes scattering entropy to detect sea ice. We tested this method by comparing sea ice area derived from PALSAR fully polarimetric data covering the Sea of Okhotsk acquired in 2009 and 2010 to that derived from reference MODIS (Moderate Resolution Imaging Spectroradiometer) data of the same region. We applied discriminant analysis to samples of sea ice and open water to determine the threshold of PALSAR backscattering coefficient and scattering entropy needed to discriminate sea ice from open water. We found that sea ice area derived from PALSAR data was equivalent to that derived from MODIS data, suggesting that our proposed method was reliable in the detection of sea ice in the Sea of Okhotsk. Index Terms Backscattering coefficient, discriminant analysis, polarimetric SAR, scattering entropy, sea ice. I. INTRODUCTION B ECAUSE sea ice acts as an insulator between air and seawater, extent of sea ice is related to local as well as global climate and can act as a critical indicator of climate change [1]. Passive microwave radiometer images play an important role in detecting changes in the size of global sea ice cover. A recent study has shown that a combination of radiometer and buoy motion data could detect changes over the last 20 years in the fraction of older, thicker ice in the Arctic Ocean [2]. Because many spaceborne synthetic aperture radar (SAR) systems currently operate in Arctic and seasonal sea ice areas, sea ice research using SAR data is of particular interest. Sea ice segmentation methods for both single- and multi-polarization SAR data have been proposed [3], [4], and a high-resolution sea ice concentration map was created for the Baltic Sea using RADARSAT ScanSAR data [5]. The relationship between ice thickness and depolarization factors was also investigated in relatively thick sea ice areas in the Arctic Sea [6]. The southern region of the Sea of Okhotsk is typically covered by ice from January to March each year. Because ice is a severe impediment for shipping, the Japan Coast Guard routinely provides a sea ice concentration map of this area spanning Manuscript received September 30, 2012; revised January 13, 2013; accepted March 21, Date of publication May 14, 2013; date of current version June 17, This work was supported by KAKENHI Grant Number The authors are with the College of Engineering, Nihon University, Koriyama, Fukushima , Japan (corresponding author hwaka@cs.ce.nihon-u.ac.jp). Color versions of one or more of the figures in this paper are available online at Digital Object Identifier /JSTARS winter months. Although they have attempted to use PALSAR (Phased-Array type L-band SAR) ScanSAR data [7] as a tool for informing sea ice concentration maps, representatives from the Coast Guard have cited difficulty in detecting thin sea ice (i.e., new ice, nilas, and grease ice) using this data because of low backscattering coefficient [8]. Detecting thin sea ice is especially important because it can prevent vessels from ingesting cooling water through a vessel s sluice gate. Wakabayashi et al. have investigated the use of Pi-SAR (airborne fully polarimetric L-band SAR) data [9], [10] for observing sea ice in the southern region of the Sea of Okhotsk. From the Pi-SAR data acquired in a wide range of incidence angles, they calculated the incidence angle dependencies of polarimetric parameters for sea ice and open water. They found that the scattering entropy could extract the open water from the sea ice region due to the low depolarization characteristics of the open water [11]. Our objective in this research was to develop a method for detecting sea ice in PALSAR data. Here, we detected sea ice area in the Sea of Okhotsk using PALSAR fully polarimetric data acquired in 2009 and 2010 and compared it to that derived from reference MODIS (Moderate Resolution Imaging Spectroradiometer) data of the same time period. II. REGION OF INTEREST AND DATA The Sea of Okhotsk is the most southerly region in the Northern Hemisphere where sea ice exists during winter months. Most sea ice found in this region has a thickness less than 1 meter. Based on ice core structure analysis, it has been suggested that the dynamical ice growth process under turbulent conditions (e.g., frazil ice formation, floe accumulation) are the dominant contributors to thick ice growth in this region [12]. Our region of interest (ROI) was located between Sakhalin and Hokkaido islands in the southern region of the Sea of Okhotsk (Fig. 1). The Japan Aerospace Exploration Agency (JAXA) launched PALSAR as a payload of the Advanced Land Observing Satellite (ALOS) in January It had been operated by April Despite frequent passes of PALSAR over the Sea of Okhotsk for the last five years, there were a few polarimetric data acquisitions available for our ROI. Four polarimetric observations were made during three consecutive winter periods from 2008 to 2010 covering our ROI. However, because of high cloud cover in our site in 2008, we focused on PALSAR and MODIS data from 2009 and Table I summarizes the list of satellite data used in this research. We used PALSAR Level 1.1 slant range complex data in order to fully extract polarimetric information. MODIS data were downloaded in /$ IEEE

2 WAKABAYASHI et al.: SEA ICE DETECTION IN THE SEA OF OKHOTSK USING PALSAR AND MODIS DATA 1517 map in the Sea of Okhotsk. However, their efforts were discontinued because of issues relating to the setting of thresholds for backscattering coefficient [8]. SARs onboard satellites launched after 2006 have the capability of acquiring polarimetric data. Fully polarimetric SAR systems transmit pulses alternatively through two antennae with different polarizations, and these antennae simultaneously receive scattered pulses from observation targets. As a result, the system can produce four polarization sets and four processed SAR images with different polarizations. Because polarimetric SAR systems observe scattering characteristics at different polarizations, they are able to gather more information regarding a target s scattering mechanisms. Fig. 1. ROI location. TABLE I LIST OF DATA USED IN THIS RESEARCH B. Scattering Entropy To solve issues highlighted in the previous section, we propose a new method for detecting sea ice using PALSAR polarimetric data. This work is based on our previous research that analyzed airborne polarimetric SAR data in the Sea of Okhotsk [11]. In this previous research, we found that the scattering entropy of open water gives consistently low values in a wide range of incidence angles because surface scattering is common in open water. Alternatively, the scattering entropy for various sea ice types generates higher values than that of open water. Therefore, we propose that scattering entropy can be used to distinguish sea ice from open water. Because the scattering entropy of thin sea ice is larger than that of thicker sea ice, we expected that this method would have an advantage in detecting thin sea ice as compared to methods based on backscattering coefficient. Scattering entropy is calculated from the eigenvalue of the covariance or coherence matrix determined from the observed scattering matrix. The scattering entropy is calculated as follows: Level 1B format from the National Snow and Ice Data Center (NSIDC) site. III. SEA ICE DETECTION USING PALSAR A. SAR Data Problems With Sea Ice Detection The backscattering coefficient of open water depends on water surface roughness, which is related to sea surface wind velocity. When water freezes, the backscattering coefficient decreases until ice thickness reaches several centimeters due to a decrease in the ice surface s dielectric constant with ice growth. In addition, because surface roughness increases as ice grows, due to the existence of frost flowers, the backscattering coefficient increases until ice thickness reaches 20 centimeters. The backscattering coefficient decreases again after ice thickness increases above 20 centimeters [13]. When a threshold for backscattering coefficient is used for sea ice detection, as is the case with single polarization SAR data, it is generally difficult to set a threshold value that differentiates sea ice and open water because of the complexity in how backscattering coefficients change as described above. The Japan Coast Guard attempted to develop a method for using PALSAR ScanSAR data to create a sea ice concentration where is the th normalized eigenvalue. Each eigenvalue is related to the ratio of different scattering mechanisms, such as odd bounce scattering, even bounce scattering, and diffuse scattering. Scattering entropy is recognized as an index of the randomness of scattering mechanisms. When scattering is predominately caused by a single scattering mechanism, scattering entropy is low. The maximum value of scattering entropy is one. Because scattering is predominately caused by surface scattering for open water, scattering entropy is low compared to that of sea ice, where scattering is the result of multiple mechanisms. Thus, we hypothesize that a threshold for scattering entropy could be used to differentiate sea ice and open water. The boundaries of these thresholds can be determined using PALSAR data. C. Processing PALSAR and MODIS Data We transformed PALSAR fully polarimetric data as well as MODIS visible and near infrared data into the same geometric coordinate systems. We processed PALSAR data according to the following steps: We calculated amplitude and scattering entropy in 12 (azimuth) by 2 (range) samples of PALSAR level 1.1 (1)

3 1518 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 6, NO. 3, JUNE 2013 Fig. 2. Example of processed images for (a) MODIS albedo on Feb. 17, 2009, (b) MODIS albedo on Feb. 20, 2010, (c) PALSAR scattering entropy on Feb. 17, 2009, and (d) PALSAR scattering entropy on Feb. 20, slant range data, which corresponded to a 50 m by 50 m ground area. Amplitude data for each polarization were converted into backscattering coefficient using calibration coefficients provided by JAXA. We transformed slant range images into UTM coordinates using cubic convolution resampling methods. We mosaicked six to seven images between Sakhalin and Hokkaido islands to create a single image of the study area. We created a land mask from SRTM-3 elevation data and applied it to the mosaicked PALSAR image. We also created a sea ice reference dataset from MODIS visible and near infrared channelsandusedittovalidateour method. We classified sea ice using standard methods based on a threshold in surface albedo: By combining three MODIS channels (bands 1, 3, and 4), the MODIS albedo was calculated according to the following equation. where,and are the reflectances calculated from digital numbers and conversion coefficients for bands 1, 3, and 4, respectively. By using thresholds in MODIS albedo, we were able to differentiate open water and to classify sea ice area into three categories (i.e., new ice, young ice and first-year ice) [14], [15]. We transformed the MODIS albedo image into UTM coordinates with 500 m pixel spacing. We applied the same land mask used for PALSAR image. We created a cloud mask from MODIS band 7 data and applied it to mask clouded areas from the MODIS image. (2) Fig. 3. Extracted areas for data analysis in (a) 2009 and (b) 2010 with (c) an example of the sampling point determination in The final product was a sea ice reference map with a spatial resolution of 500 m. Examples of PALSAR scattering entropy and MODIS albedo images are given in Fig. 2. The size of MODIS images was 1000 by 1000 pixels with 500 m pixel spacing. D. Extraction of Sea Ice and Open Water Areas Because PALSAR and MODIS images were geographically aligned, we could extract the characteristics of PALSAR data that corresponded to various MODIS albedo including sea ice and open water. However, because of an eight hour time difference in acquisition time between PALSAR and MODIS, extraction errors caused by sea ice drifts during this time lag had to be carefully considered. We visually evaluated errors in sea ice position and excluded those areas from our analysis. We ultimately confined all further analysis to regions within our study area where sea ice movement was limited.

4 WAKABAYASHI et al.: SEA ICE DETECTION IN THE SEA OF OKHOTSK USING PALSAR AND MODIS DATA 1519 Fig. 4. Relationship between MODIS albedo and (a) PALSAR scattering entropy, (b) HH backscattering coefficient, (c) VV backscattering coefficient, and (d) HV backscattering coefficient in 60 sea ice and 20 open water areas. Each dot indicates an averaged value in 2.5 km by 2.5 km area. Fig. 3 (a) and (b) indicate the areas we analyzed for both sea ice and open water overlaid on PALSAR scattering entropy data. Because MODIS albedo changes with varying sea ice types, we extracted sea ice areas with a MODIS albedo from 0.1 to 0.3. We extracted 10 and 30 areas for open water and sea ice analysis, respectively, from the 2009 and 2010 datasets, each with an area of 2.5 km by 2.5 km. Fig. 3(c) shows enlarged MODIS and PALSAR images for the southern offshore region of Sakhalin Island where the boundary between sea ice and open water can be recognized in both images. Because of boundary changes between the two images, we purposely selected sampling areas in this region that were distant from boundaries. In total, we selected 20 open water and 60 sea ice areas for analysis. In cases where drifting errors were present within the analyzed area, errors in extracted backscattering coefficient or scattering entropy may have occurred. According to scattergrams, scattering entropy could be used to differentiate open water and sea ice where backscattering coefficients could not (Fig. 4). We observed a bimodal separation in scattering entropy for open water and sea ice (Fig. 5). IV. QUANTATATIVE ANALYSIS FOR DISCRIMINATION OF SEA ICE AND OPEN WATER We quantitatively verified detection accuracy of sea ice and open water using multivariate discriminant analysis [16]. We first determined within-class and between-class variance of two classes, sea ice and open water, which was recognized in MODIS albedo. A correlation ratio was then calculated as the ratio of between-class variance to total variance, which was used as an index of separation between classes. From within-class variance and between-class variance, the total variance was calculated as. Finally, the correlation ratio was defined as: We applied a linear discriminant analysis (LDA) in order to separate sea ice and open water by backscattering coefficients and scattering entropy, ultimately determining a threshold separatingthetwoclasses.throughlda,aseparationline (3)

5 1520 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 6, NO. 3, JUNE 2013 Fig. 5. Distribution of scattering entropy and backscattering coefficients. Since the unit area in this analysis is 500 m by 500 m, these are the histograms of 2000 data for (a) Scattering entropy, (b) HH backscattering coefficient, (c) VV backscattering coefficient, and (d) HV backscattering coefficient. between two classes can be established through the following relationship: TABLE II CORRELATION RATIOS AND DISCRIMINANT ACCURACIES BETWEEN SEA ICE AND OPEN WATER (4) where is a new variable, istheoriginalvariable(here,either backscattering coefficient or scattering entropy), and and are constants chosen to maximize the correlation ratio of the two classes. The other conditions in determining and are the variance of to be number of samples, and the center between two class averages to be zero. Because the two classes are separated at zero in, the threshold in the original variable is determined as. Table II summarizes the resulting correlation ratios, derived thresholds for open water and sea ice, and discriminant accuracies, which were derived from 2000 samples of backscattering coefficients and scattering entropy. We found that polarimetric SAR data could be used to effectively discriminate sea ice and open water; the correlation ratio for scattering entropy was 0.73, which was larger than the ratio for backscattering coefficient. Here, because VV polarization had the highest correlation ratios within all polarizations, it was superior to HH polarization in discriminating these two classes. The backscattering coefficient for HV polarization alone did not accurately separate sea ice and open water. Discriminant accuracies of the derived thresholds were highest for scattering entropy followed by VV, HH, and HV polarizations. We also investigated combinations of multiple polarization sets in addition to the single polarization backscattering coeffi-

6 WAKABAYASHI et al.: SEA ICE DETECTION IN THE SEA OF OKHOTSK USING PALSAR AND MODIS DATA 1521 TABLE III CORRELATION RATIOS AND DISCRIMINANT ACCURACIES FOLLOWING LINEAR DISCRIMINANT ANALYSIS USING BACKSCATTERING COEFFICIENTS cient. The relationship for new variables that combined multiple polarizations was as follows: (5) where is the th polarization backscattering coefficient, and are coefficients that maximize the correlation ratio of two classes, and is the number of polarizations in the combination. Table III summarizes the resulting correlation ratios and discriminant accuracies for the combination of multiple polarization backscattering coefficients. We observed high correlation ratios and discriminant accuracies for combinations of like- and cross-polarizations in a dual polarization combination. The VV and HV combination had the highest correlation ratio (0.706) with discriminant accuracy (98.4%). We also investigated the combination of three polarizations (HH, VV, and HV) and found a similar correlation ratio (0.718) and discriminant accuracy (99.3%). Thus, the best polarization combination for discriminating sea ice and open water was the VV and HV combination for this dataset. V. DISCUSSION A. Backscattering Characteristics of Open Water A total sample number of open water recognized by MODIS albedo was 500, and the mean backscattering coefficients were db for HH, db for VV, and db for HV. The range of incidence angle was 23 to 25 degrees for the PALSAR 21.5 degrees off-nadir angle. Mean wind speed at PALSAR acquisition time (12:30 UT) was 3.2 m/s and 4.6 m/s for the 2009 and 2010 datasets, respectively, which were recorded at Tokoro AMeDAS (Automated Meteorological Data Acquisition System) station at the southern edge of our ROI on Hokkaido Island [17]. Based on the relationship between PALSAR HH backscattering coefficient and wind speed as investigated by Isoguchi et al. [18], we estimated the backscattering coefficient at approximately to db over our ROI. Because the estimated backscattering coefficients were smaller than the measured values, the actual wind speed over offshore areas was assumed to be larger than the measured wind speed at Tokoro station. Considering monthly average wind speed in February at Tokoro AMeDAS station, which was 4.4 m/s in 2009 and 3.5 m/s in 2010, the sea surface condition at PALSAR acquisition time was considered to be typical for this time of year in our ROI. In theory, when the ocean surface becomes specular due to no wind, the backscattering coefficient decreases and approaches Fig. 6. Enlarged MODIS and PALSAR images for possible areas of thin sea ice (indicated rectangle on (a)), including (b) MODIS albedo, (c) PALSAR HH backscattering coefficient, and (d) PALSAR scattering entropy. The size of these images is 34 km by 34 km. the noise level. In this case, it is possible that open water would not accurately be differentiated because of increased scattering entropy. However, we did not find that this situation occurred in areas classified as open water in the 2009 and 2010 MODIS data. We plan to further investigate this phenomenon using various SAR systems, (e.g., ALOS-2/PALSAR-2), which will be available in the near future. B. Characteristics and Detection of Thin Sea Ice The average backscattering coefficients for 1500 sea ice samples were db for HH, db for VV, and db for HV. Considering the backscattering coefficients and observation incidence angles, we concluded that these values were not relevant to thin sea ice with thickness less than 10 cm. To determine if our method could accurately detect thin sea ice, we examined the entire 2009 and 2010 dataset. We found possible areas of thin sea ice in the offshore region of the east coast of Sakhalin Island. The enlarged area was shown in Fig. 6, and the characteristics of this area were summarized in Table IV. We could recognize the areas with resemble size and shape in both MODIS and PALSAR images in Fig. 6. Because the displacement between MODIS and PALSAR images in this area was relatively large, we did not include the characteristics of this area in our quantitative analysis described in Section IV. Wakabayashi et al. [11] showed that thin sea ice had backscattering coefficients less than db and scattering entropies larger than 0.4 in the range of incidence angles from 20 to 30 degrees. We found similar values for thin sea ice in our ROI (Table IV), and we conclude that scattering entropy can be used to accurately detect thin sea ice areas even in areas of low backscattering. Although it is possible to detect thin sea

7 1522 IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, VOL. 6, NO. 3, JUNE 2013 TABLE IV CHARACTERISTICS OF POSSIBLE AREAS OF THIN SEA ICE VI. SUMMARY In this study, we proposed a new method for detecting sea ice in the Sea of Okhotsk by using PALSAR fully polarimetric data. We investigated backscattering coefficients as well as scattering entropy behaviors for areas with sea ice and open water, which were recognized using MODIS visible and near infrared data. After analyzing these two classes with linear discriminant analysis, we found that sea ice and open water areas were best differentiated using scattering entropy compared to any combination of backscattering coefficients. Using the distribution of scattering entropy and backscattering coefficients for sea ice and open water, thresholds for scattering entropy and backscattering coefficients were determined within our ROI in the Sea of Okhotsk. Our results demonstrated the feasibility of sea ice detection in the Sea of Okhotsk using L-band fully polarimetric SAR data. In the future, we plan to further investigate methods for detecting sea ice with various conditions using additional SAR, including ALOS-2/PALSAR-2. ACKNOWLEDGMENT This work was partly conducted under the agreement of JAXA Research Announcement titled Sea ice study and its application using PALSAR polarimetric data in the Sea of Okhotsk (JAXA-PI: 205). The PALSAR data were distributed under this agreement. The Arctic Research using IARC-JAXA Information System (IJIS) partly supported this research. Fig. 7. Example of a sea ice concentration map derived from PALSAR data from (a) 2009 and (b) The spatial resolution of these maps is 500 m by 500 m. ice area by locating areas of low backscattering, we suggest that using scattering entropy together with backscattering coefficient is a more reliable way of detecting thin sea ice. We plan to further analyze the application of this methodology using simultaneously acquired SAR (e.g., ALOS-2/PALSAR-2) and optical sensor data. C. Sea Ice Concentration Derived by PALSAR Data Sea ice concentration is defined as the ratio of area covered by ice to the total area for a given sea area. A high-resolution map of sea ice concentration would be important for ship navigation, which could be constructed from SAR data. We created a high resolution binary image of sea ice by setting a threshold of 0.25 on the PALSAR image of scattering entropy. Based on the definition of sea ice concentration, we calculated the ratio of pixels classified as sea ice to the total number of pixels in the image. Fig. 7 shows an example of a sea ice concentration map created from PALSAR data acquired in 2009 and REFERENCES [1] F. Nishio and M. Aota, Variability of sea ice extent in the Sea of Okhotsk, in Proc. Int. Symp. ISY Polar Ice Extent, Hokkaido, Japan, 1993, pp [2]M.Tschudi,C.Fowler,J.Maslanik,andJ.Stroeve, Trackingthe movement and changing surface characteristics of Arctic sea ice, IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol.3,no.4, pp , [3] X. Yang and D. A. Clausi, Evaluating SAR sea ice image segmentation using edge-preserving region-based MRFs, IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 5, no. 5, pp , [4] A. P. Doulgeris, S. N. Anfinsen, and T. Eltoft, Automated non-gaussian clustering of polarimetric synthetic aperture radar images, IEEE Trans. Geosci. Remote Sens., vol. 49, no. 10, pp , [5] J. Karvonen, Baltic sea ice concentration estimation based on C-band HH-polarized SAR data, IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 5, no. 6, pp , [6] J. W. Kim, D. J. Kim, and B. J. Hwang, Characterization of Arctic sea ice thickness using high-resolution spaceborne polarimetric SAR data, IEEE Trans. Geosci. Remote Sens., vol. 50, no. 1, pp , [7] A. Rosenqvist, M. Shimada, N. Ito, and M. Watanabe, ALOS PALSAR: A pathfinder mission for global-scale monitoring of the environment, IEEE Trans. Geosci. Remote Sens., vol. 45, no. 11, pp , [8] S. Fukushima et al., Joint research on the best use and the development of the sea ice observation method using ALOS data, Kaiho Report, vol. 26, pp , 2008, (in Japanese). [9] T. Kobayashi, T. Umehara, M. Satake, A. Nadai, S. Uratsuka, T. Manabe, H. Masuko, M. Shimada, H. Shinohara, H. Tozuka, and M. Miyawaki, Airborne dual-frequency polarimetric and interferometric SAR, IEICE Trans. Commun., vol. E83-B, pp , [10] T. Matsuoka, S. Uratsuka, M. Satake, A. Nadai, T. Umehara, H. Maeno, H. Wakabayashi, and F. Nishio, CRL/NASDA airborne SAR (Pi-SAR) observations of sea ice in the Sea of Okhotsk, Ann. Glaciol., vol. 33, pp , 2001.

8 WAKABAYASHI et al.: SEA ICE DETECTION IN THE SEA OF OKHOTSK USING PALSAR AND MODIS DATA 1523 [11] H. Wakabayashi, T. Matsuoka, K. Nakamura, and F. Nishio, Polarimetric characteristics of sea ice in the Sea of Okhotsk observed by airborne L-band SAR, IEEE Trans. Geosci. Remote Sens., vol. 42, no. 11, pp , [12] T. Toyota, K. Baba, E. Hashiya, and K. I. Ohshima, In-situ ice and meteorological observations in the southern Sea of Khotsk in 2001 winter: Ice structure, snow on ice, surface temperature, and optical environments, Polar Meteorol. Glaciol., vol. 16, pp , [13] R.G.Onstott,,F.D.Carsey,Ed., SARandscatterometersignaturesof sea ice, in Microwave Remote Sensing of Sea Ice (Geophysical Monograph 68). Washington, D.C.: American Geophysical Union, 1992, pp , in. [14] J. F. Heinrichs, D. J. Cavalieri, and T. Markus, Assessment of the AMSR-E sea ice concentration product at the ice edge using RADARSAT-1 and MODIS imagery, IEEE Trans. Geosci. Remote Sens., vol. 44, no. 11, pp , [15] D. K. Hall, D. J. Cavalieri, and T. Markus, Assessment of AMSR-E antarctic winter sea-ice concentrations using aqua MODIS, IEEE Trans. Geosci. Remote Sens., vol. 48, no. 9, pp , [16] R. A. Fisher, The use of multiple measurements in taxonomic problems, Annals of Eugenics, vol. 7, no. 2, pp , [17] F. Fujibe, Long-term changes in wind speed observed at AMeDAS stations, Tenki, vol. 50, no. 6, pp , 2003, (in Japanese). [18] O. Isoguchi and M. Shimada, An L-band ocean geophysical model function derived from PALSAR, IEEE Trans. Geosci. Remote Sens., vol. 47, no. 7, pp , Hiroyuki Wakabayashi (M 99) received the B.S. and M.S. degrees in electronic engineering and the Ph.D. degree in applied physics, all from Hokkaido University, Hokkaido, Japan, in 1981, 1983, and 1996, respectively. In 1983, he joined the National Space Development Agency in Japan (NASDA, currently JAXA). From 1986 to 1991, he was in charge of developing data processing and Cal/Val systems for JERS-1. From 1994 to 1999, he led the ALOS/PALSAR conceptual design. From 1999 to 2006, he was a Research Scientist at Earth Observation Research Center and engaged in polarimetric SAR research. From 2001 to 2003, he was a member of the 43rd Japanese Antarctic Research Expedition (JARE) and stayed at the Syowa Station, where he was in charge of SAR data processing and analysis for sea ice monitoring. Since April 2006, he has been with the College of Engineering, Nihon University, where he is currently a Professor with the Department of Computer Science. Dr. Wakabayashi is a member of IEICE, the Remote Sensing Society of Japan, the Japan Society of Photogrammetry and Remote Sensing, and the Japanese Society of Snow and Ice. Yuta Mori was born in Fukushima, Japan, in He received the B.E. degree in computer science from the College of Engineering, Nihon University, Japan, in 2011, where he is currently working on the M.E. degree in computer science. His research interests include radar and field data analysis for sea ice monitoring. Kazuki Nakamura was a Postdoctoral Fellow in remote sensing science with the Communications Research Laboratory (CRL) (currently the National Institute of Information and Communications Technology: NICT), Koganei, Tokyo, Japan, from 2003 to 2006, and the National Institute for Polar Research (NIPR), Itabashi, Tokyo (currently Tachikawa, Tokyo), Japan, from 2006 to In , he was a Postdoctoral Fellow in remote sensing science with the National Institute of Advanced Industrial Science and Technology (AIST), Tsukuba, Ibaraki, Japan. Since 2012, he has been an Associate Professor with the Department of Computer Science, College of Engineering, Nihon University, Koriyama, Fukushima, Japan. His research interests are in microwave remote sensing of sea-ice, glacier and various targets of the earth.