Outliers can be hard to spot since the residuals from one or several outliers tend to be distributed over all residuals. The problem is particularly large when dealing with automatic picks (A in column 16 of phase line) where gross mistakes, like picking P at S, can occur. One possibility is to use residual weighting but often more drastic means are needed. Fernando Carrilho (email@example.com) has made an algorithm for HYP which effectively finds and remove and/or weight out large outliers. Only P and S-phases will be checked. The procedure works for both local and distant events although it is mainly meant to be used with local events. The principle is the following:
Run the location in a loop (iterations) of a maximum of 15 steps
For each step:
- Weight out residuals larger then 99
-S-phase: If residual larger than 1.5*rms and larger than RESET TEST(109) it is weighted out. If in a following iteration it is smaller than 1.5*rms, and smaller then RESET TEST(109), it is included again. If 1.5*rms is less than 0.6 and the residual is less than 1.0, the data is always used. Default value of RESET TEST(109) is 0.8.
-P-phase: Same as for the S-phase but only after the 2. iteration.
The loop will be terminated if any of 2 conditions are met:
-rms is less than 0.3 and iteration is more then 5.
-Change in rms from one step to the next is less than 0.05 and iteration is larger than 4.
The phases with high residuals are now weighted out. They can also be removed from the S-file if the residuals are larger than a user defined value given in RESET TEST(110) (default value of 3 ), however this is only possible for the phases flagged as automatic.
There is also an option for keeping only the phase with the lowest residual if the same phase has been picked on several components of the same station. Phases will be deleted for both manual and automatic phase picks. Set RESET TEST(111)=1.0. Default value is 1.0.
There is an option for testing for Sg phases. If RESET TEST(112) is set to 1.0, there will be an additional iteration loop, where all S- phases which are only labeled S have been changed to Sg. At the end, the data set with the highest number of accepted phases will be used and all S's will be S or Sg. The phases which already have two letters like SG or SN will not be changed. This option will not be used for distant or regional events.
The default values are the ones best fitting automatic operation using AUTOPIC in the Portuguese network. Increasing the values of test(109) and test(110) will increase the number of accepted phases.