This event is quite small for teleseismic stations, which means it will be more strongly affected by the 3D structure at the ray path.
I recommend you to use local stations (<500km) only if the azimuth covourage is good, and test various models to generate corresponding Green's functions, it seems to me that your filtering pass is quite long period, you can try to use some higher frequency pass to catch more information
Meanwhile, it seems some of the stations have good fit, you can give them larger weighting factors in weight.dat file and discard bad traces