We made the same analysis for the other 8 CSST stations(not shown here).Figures 3(a)through(b)show the sumnrary of comparison for 9 CSST stations.Figure 3(a)shows the correlation coeffieients between CSST and SST anomalies at each station.The brackets are attached to the cases which do not exceed the 98% significant level in the Student t-test for correlation coefficient.Here,we estimate the degrees of freedom using the data number(total months used in comparison)divided by the integral time scale of 5 months based on estimation from the auto-correlation analysis to individual time series(e.g.,Davis,1976).It is seen four stations(Stations d,e,g,and h)among 9 stations have cases in which correlation coefficients do not exceed the 98% significant level.We regard these stations are not appropriate for comparison,and therefore the results of these four stations are excluded in the later consideration.Figure 3(a)shows that in general correlation coefficients for the period of 1942-1971 are greater than those for the period of 1912-1941.This may reflect the relatively low number of SST data in the period of 1912-1941 compared with that in the period of 1942-1971.
Figure 3(b)shows mean deviations of uncorrected SST,corrected SST and CSST in the period of 1912-1941 at each station.As mentioned before,if the present bucket correction is perfect,that is,it functions perfectly,then the difference using corrected SST(closed triangles)should coincide with that of CSST(closed stars).We can see that at Stations c,f and i those are almost same,actually those differences are 0.03℃,0.10℃ and 0.00℃,respectively.At Stations a and b,differences between mean deviation of corrected SST and that of CSST amount to 0.25℃ and 0.18℃,respectively.However,these numbers are fairly improved compared with 0.41℃ and 0.36℃ between that of uncorrected SST and that of CSST.
4. Summary
In the present study,we have examined the validity of correction for historical SST data taken by the bucket method,which was proposed by Folland and Parker(1995),using the long-term CSST data taken at 9 stations around Japan.As a result,the data of 5 CSST stations among 9 stations were suitable for comparison.It can be concluded that the correction to the historical SST data proposed by Folland and Parker(1995)functioned very well in the historical SST data computed from COADS and the Kobe Collection,at least to the sea areas around Japan;That is,the systematic biases in monthly mean SST anomalies have been corrected almost perfectly at three stations,and the biases at the other two stations have been reduced by 40-50%.Precisely speaking,since only the CSST data taken around Japan were used,we can not say anything on the validity of this correction in the other sea areas.If another appropriate long-term CSST data are available in the world's oceans,the same comparison will be useful,and the validation of the present correction will be able to be confirmed much more robustly.Further,it can be pointed out that the present method using the coastal meteorological stations as well as CSST stations can be applied for another marine elements such as sea level pressure and so on.
