To estimate Qscs, previous studies used the spectral ratios of multiple ScS phases calculated for specific time windows. They assumed that the spectral ratios had linear relation with frequency. However, the spectral ratios very often did not seem to be linear, and the error bars were very large for frequencies higher than about 0.05 Hz.
We have performed numerical experiments to search for the cause of these anomalies and found that they are caused by the multiple reflection in the crust under the surface bounce point of ScS2. A method of suppressing the anomalies is to start the time of the time window at 40 sec prior to the theoretical arrival times of the multiple ScS phases. In numerical tests, the difference between the correct and estimated values of Qscs is only 0 to 5% using this method, but it is 10 to 50% in the conventional method. We find that the crustal effect is more important in the estimation of Qscs than background noise, which has been readily handled in previous studies using standard stacking techniques.