I'm trying to pick the P onset of an event with magnitude Mw6.3. The waveforms at tel-stations seems to have a relative large noise which makes trouble for me. The rupture duration is about 2s and I have nearly 1000 traces. So could you give me some advice on how to pick them accurately and efficiently(I hope the error is no more than 0.1~0.2s) like stacking or something else. Some automatically method is preferred. I have tried cross-correlation, however the waveform changes a lot for the whole data and it fails easily.