10:45 AM - 12:15 PM
[ACG46-P07] Validation of AMSR2 thin ice thickness algorithm for global sea-ice covered oceans
Keywords:Thin ice thickness, AMSR2, Coastal polynya, Frazil ice
Coastal polynya (thin ice areas and low ice concentration areas) are areas of high ice production due to huge heat loss to the atmosphere. Thin ice thickness algorithms that use satellite passive microwave data have been developed to detect coastal polynyas and estimate thin ice thickness on a daily timescale. In these algorithms, ice thickness of <20 cm is empirically estimated based on a negative correlation relationship between the polarization ratio (PR) of brightness temperatures (TBs) and ice thickness (hi). To improve the accuracy of ice thickness estimates, a recent thin ice thickness algorithm by Nakata et al. (2019) first classifies thin ice areas (polynyas) into two ice types: active frazil, which is a frazil/grease ice area formed under turbulent conditions, and thin solid ice, which is a relatively uniform thin ice such as nilas under calm conditions. Then, thin ice thickness is estimated from the PR-thickness relationship for each ice type. In this study, we have examined whether this algorithm can apply to data from AMSR2 which is the successor of AMSR-E, for the global sea-ice-covered oceans. The footprint size of AMSR2 is improved by about 85% from AMSR-E despite the frequency channels being the same. By adding AMSR2 data, thin-ice thickness, ice type, and ice production data set with a higher spatial resolution can be obtained from 2002 to the present.
Synthetic aperture radar (SAR) images are capable of distinguishing between active frazil and solid ice. In this study, four Sentinel-1A images obtained around Antarctica were used to validate the ice type classification. Ice thickness for comparison is derived from heat flux calculations that use atmospheric data and ice surface temperatures from Aqua/MODIS thermal infrared images (Band 31 and 32). We used ~230 cloud-free MODIS images collected from the the Sea of Okhotsk, Bering Sea, Chukchi Sea, North of Baffin Bay, Ross Sea, and oceans around East Antarctica. As the atmospheric input data, we used air temperature at 2 m, dewpoint temperature at 2 m, wind speed at 10 m, and surface sea level pressure from 1-h ECMWF ERA5 reanalysis data. The brightness temperatures at 36 GHz and 89 GHz, which are the input data for the thin ice algorithm, are those from the AMSR2 Level 1R product.
We obtained the correct data for the sea ice type from the visual inspection of Sentinel-1A images. The comparison of that data with the AMSR2 classification results indicates that the active frazil areas can be identified by our algorithm, with the misclassification of ~2%. Since AMSR2 PR-hi relationships slightly differ from those for AMSR-E in Nakata et al. (2019), we estimate ice thickness using newly derived empirical curves that best fit into the AMSR2 data. And then, AMSR2 ice thicknesses were compared with those estimated from MODIS data. The Mean absolute errors are calculated to be 1.3 and 4.7 cm for active frazil and thin solid ice, respectively, which meet the target accuracies (active frazil:3 cm, thin solid ice: 10 cm) for release as a JAXA research product.
Synthetic aperture radar (SAR) images are capable of distinguishing between active frazil and solid ice. In this study, four Sentinel-1A images obtained around Antarctica were used to validate the ice type classification. Ice thickness for comparison is derived from heat flux calculations that use atmospheric data and ice surface temperatures from Aqua/MODIS thermal infrared images (Band 31 and 32). We used ~230 cloud-free MODIS images collected from the the Sea of Okhotsk, Bering Sea, Chukchi Sea, North of Baffin Bay, Ross Sea, and oceans around East Antarctica. As the atmospheric input data, we used air temperature at 2 m, dewpoint temperature at 2 m, wind speed at 10 m, and surface sea level pressure from 1-h ECMWF ERA5 reanalysis data. The brightness temperatures at 36 GHz and 89 GHz, which are the input data for the thin ice algorithm, are those from the AMSR2 Level 1R product.
We obtained the correct data for the sea ice type from the visual inspection of Sentinel-1A images. The comparison of that data with the AMSR2 classification results indicates that the active frazil areas can be identified by our algorithm, with the misclassification of ~2%. Since AMSR2 PR-hi relationships slightly differ from those for AMSR-E in Nakata et al. (2019), we estimate ice thickness using newly derived empirical curves that best fit into the AMSR2 data. And then, AMSR2 ice thicknesses were compared with those estimated from MODIS data. The Mean absolute errors are calculated to be 1.3 and 4.7 cm for active frazil and thin solid ice, respectively, which meet the target accuracies (active frazil:3 cm, thin solid ice: 10 cm) for release as a JAXA research product.