Three-dimensional empirical traveltimes: Construction and applications

A new approach for predicting traveltimes is proposed which makes use of sophisticated smoothing and interpolation techniques applied to observations from large earthquake catalogues. The technique produces 3-D empirical traveltimes directly from a database of observations and does not impose any arbitrary spatial scalelength on the data, for example by smoothing observations over a pre-determined distance interval. This feature allows the spatial variability in data density, as well as multiple length scales of heterogeneity, to be handled in a natural way. After removing the empirical traveltimes from observed traveltimes, the remaining residuals are spatially uncorrelated and, in this sense, all of the heterogeneity signal in the data is accounted for. Traveltimes predicted with the new approach compare favourably with those from 1-D and 3-D earth models. Application to nuclear blast data shows that traveltime residuals are reduced by 62 per cent, and for earthquake residuals a reduction of up to 42 per cent is achieved across a range of teleseismic phases. Our results show that the pattern of regional variability in P-wave traveltimes can also be recovered accurately. In the epicentre determination of 25 nuclear blasts, the use of empirical traveltimes in place of 3-D global velocity models reduced the average mislocation by over 59 per cent. This new approach can also be used to 'de-noise' large arrival time databases and to identify outliers. The methodology is quite flexible and can equally well be applied to local and regional seismicity.