ON THE ONSET OF P-SEISMOGRAMS

For more than 30 years, seismologists have been encouraged to read P o nsets to 0.1 s. Evidence is presented that the actual errors are signi ficantly greater than this. Three main factors determine how accuratel y any particular onset is read: (1) magnification and timescale of the display; (2) signal-to-noise ratio; and (3) the form of the P pulse r adiated by the source. If too low a magnification is used, the first m otion can have such a low amplitude that it is not identified and the reported onset is measured on a later arrival. Potentially, the larges t errors will occur for stations for which direct P leaves the source near a node in the earthquake radiation pattern. If the signal-to-nois e ratio is low, weak nodal P may not be observed and the onset picked on a later large-amplitude phase such as the surface reflection pP. Th e size of the first motion relative to the maximum amplitude of P depe nds principally on the rise time of the P pulse. For explosions where the rise time can be very short, first motion on conventional SP seism ographs may be more than one-quarter of the maximum peak-to-trough def lection shown by P. For some earthquakes where the rise time is over 0 .5 s, however, first motion may be only 1/20 of the maximum peak-to-tr ough deflection and so may be lost in the noise. Examples are shown of seismograms that display such features. Estimates of the reading erro r in P have been determined using the joint epicentre method to estima te the relative epicentres and origin times of groups of explosions at two test sites (Degelen Mt and Shagan River in Kazakhstan) and for a series of aftershocks (the 1976 Gazli, Uzbekistan sequence). As statio n terms take up deviations from standard traveltimes that are common t o all disturbances in a group, the residuals after applying the joint epicentre method should be a measure of observational error, assuming that the effects of any lateral heterogeneity in the source region are negligible. The joint epicentre studies show, using P onsets measured on array seismograms, that the standard deviations of the residuals ( on seven degrees of freedom) for the Degelen Mt and Shagan River explo sions are 0.10 s and 0.14 s respectively, whereas for the Gazli earthq uakes the standard deviation is 0.39 s, the difference between the var iances for the earthquakes and explosions being significant at <0.1 pe r cent level. The results show that even for the onsets from explosion P waves recorded at arrays, which are some of the most reliable data available, observational errors are not consistently less than 0.1 s a nd for earthquakes the errors will often be 0.5 s or greater.