By using the brightness tempereture data collected by NOAA-9 AVHRR and the sea surface temperature (SST) data by Mutsu Bay automatic marine monitoring buoy system, the SST estimation accuracy by the split window function (SWF) method was validated. From the archived AVHRR images collected during 1986-1988, a validation data set of 390 match-ups was constructed after a rigid cloud-free and noise-free screening. The temporal and spatial coincidence in each match-up is within 30 minutes and 1 pixel resolution.
A split window function for the SST estimation was derived by applying the regression analysis. Its root mean of the squared errors (RMSE) was 0.59^°C. There appeared match-ups with rather large errors. By referring to the meteorological records of various items at their data collections, it is found that large errors tended to appear when large differences existed between the air temperature and the sea surface temperature. The main reason of the large errors was confirmed due to the sea surface effects, i.e., the vertical water temperature distribution just near the sea surface. The satellite detects the skin SST, but the buoy measures the bulk SST at 1 m below the sea surface. By removing the match-ups with large errors from the original data set, a selected data set having 334 match-ups was prepared and its RMSE was 0.34^°C.