Validation And Intercomparison Of Multi-sensor Satellite Sea Surface Temperature Data

Remote Sensing techniques have the advantage of synoptic and frequent view of the large geographic areas, which is useful and convenient for the studies of the global oceans. The accuracy of the Sea Surface Temperature (SST) derived from satellite measurements is one of the key factors of climate prediction. Obviously, it is necessary to validate the satellite SSTs derived from multi-sensors and compare different SST datasets. The accuracy of SST datasets will be improved, which will promote the operational applications of SST datasets; on the other hand, the results will be very important for data merging and climate analysis. The studies on satellite SST validation and intercomparison published in recent years is investigated. Validation and intercomparison of multi-sensor SST datasets in the Northwest Pacific during the period of 2005 are carried out. After the preprocessing of data, quality control and generation of mach-up database, the consistency and characteristics of the differences between the satellite SST datasets and in situ buoy SST data are compared and analyzed. The characteristics of the differences among different SST datasets and error sources are also analyzed.Firstly, multi-sensor Satellite SST datasets in the Northwest Pacific are compared with in situ buoy SST data respectively. MCSST and NLSST are two products with different retrieval algorithms, which are received and processed by the Satellite Ground Station of Ocean University of China. The results of preliminary validation show that MCSST and NLSST data are lower than in situ buoy SST data. The annual mean bias of the difference between in situ buoy SST data and MCSST data in day and night time is 0.23°C and 0.36°C respectively, and the annual standard deviation of it in day and night time is 0.88°C and 0.93°C respectively; The mean bias of the difference between in situ buoy SST data and NLSST data in day and night is 0.04°C and 0.25°C respectively. It has a small change in the annual standard deviation in day and night which is 0.87°C and 0.93°C respectively. Matchup data with satellite zenith angle greater than 45°are excluded,the mean bias of the difference between in situ buoy SST data and NLSST in day and night is -0.16°C and 0.13°C respectively; It has a good change in the annual standard deviation in day and night which is 0.73°C and 0.80°C respectively. The results indicate that NLSST data is better than MCSST data,so it is used in intercomparisons of the multi-sensor satellite SST datasets。. The four satellite SST datasets of NOAA/NASA AVHRR Pathfinder (PFSST), Aqua/MODIS SST, Terra/MODIS SST, and AMSR-E SST are compared with in situ buoy SST data too. The annual mean bias of the difference between in situ buoy SST data and these datasets is -0.08°C/-0.08°C (day/night, as follows), -0.12°C/0.34°C, 0.00°C/0.11°C, and -0.20°C/-0.18°C respectively. The annual standard deviation of them is 0.60°C/0.68°C, 0.67°C/0.76°, 0.69°C/0.74°C, and 0.64°C/0.67°C respectively. The statistics shows that the annual mean bias is small except that of MCSST and that of AquaSST in night time. Generally, AMSR-E SST is higher than in situ buoy SST.Secondly, the multi-sensor satellite SST datasets are intercompared and the characteristics of the differences between them are analyzed. The intercomparisons of multi-sensor SST data with NLSST data show that NLSST is lower than other dataset in night,and it is higher than the other infrared dataset. The annual mean bias between AquaSST and NLSST is lower while that between AMSR-E and it is the largest. The results of the intercomparisons also show AMSR-E SST is higher than the infrared SST datasets. The consistency of PFSST and MODIS SSTs is the best except for AquaSST in night and the annual standard deviation and mean bias are low. The statistics of the differences show regional and seasonal variations. The error sources may come from different measuring principles, different retrieval algorithms, different resolutions and different atmospheric conditions.In summary, the statistics shows that different SST datasets are consistent with each other. Some infrared SST data are contaminated by cloud. Cloud detection and the retrieval coefficients of AVHRR LAC SST are needed to be improved. Correction of multi-sensor satellite SST datasets should be made for the applications required high accuracy SST such as study of climate change and data merging.