NASA TEAM ALGORITHM FOR SEA ICE CONCENTRATION RETRIEVAL FROM DEFENSE METEOROLOGICAL SATELLITE PROGRAM SPECIAL SENSOR MICROWAVE IMAGER - COMPARISON WITH LANDSAT SATELLITE IMAGERY

Validation of the NASA team algorithm for the determination of sea ice concentrations from the Defense Meteorological Satellite Program special sensor microwave imager (SSM/I) is described. A total of 28 cloud-free Landsat scenes were selected in order to permit validation of the passive microwave ice concentration algorithm for a range of ice concentrations and ice types. The sensitivity of the NASA team algorithm to the selection of locally and seasonally adjusted algorithm parameters is discussed in detail. Mean absolute differences between SSM/I and Landsat ice concentrations are within 1% during fall using local and global tie points. Standard deviations of the difference are +/- 3.1% and +/- 6.2% respectively. The overall accuracy of the NASA team algorithm is lower in spring than in fall. In areas with greater amounts of nilas and young ice, we found that the NASA team algorithm underestimates ice concentrations by as much as 9%. The Landsat and SSM/I ice concentrations have a correlation of 0.968 for all spring and fall case studies with a standard deviation of +/- 6.6% using global tie points and a correlation of 0.982 with a standard deviation of +/- 4.5% using local tie points. The NASA team algorithm tends to underestimate ice concentration in areas of close pack ice and to overestimate ice concentrations in areas of open pack ice. In summer, mean differences between SSM/I and Landsat ice concentrations are 3.8% for local tie points and 11.0% for global tie points for Arctic areas and 7.2% for local tie points and 11.7% for global tie points for Antarctic areas. These large differences are attributable to surface melt during summer and comparison problems arising from a time lag of up to 8 hours between the DMSP and Landsat satellites. It appears that seasonally and regionally adjusted tie points (local tie points) will improve the overall performance of the NASA team algorithm. Our work suggests that the standard deviation between SSM/I and Landsat ice concentrations decreases from +/- 7% to +/- 5% with local tie points compared to global ones for spring and fall.