We validate the performance of three P‐velocity models built with different approaches on regional travel‐time prediction for the Tethyan margin in order to test how well they predict independently observed travel times. The three models are constructed with travel‐time tomography, a compilation of a priori geologic and geophysical information, and empirical scaling with adjustment from P‐arrival inversion, respectively. We compared the synthetics with reference travel times (ground truth data) obtained by using events or explosions located within 25 km with 95% confidence. We found variance of travel times is not an adequate tool to assess the performance of velocity models, because predicted travel times that have small variance can have very different mean value from that of observed ones. Therefore, we propose an alternative way, variance estimation with mean of observed travel times (zero mean). This technique is more efficient to assess the mismatch between synthetics and observed travel times. Among the three models we investigated, the EAPV11 model built mainly with the empirical scaling shows better performance on the travel‐time prediction. This result is intriguing, because this model inherits crustal velocity structure, Moho depth, Pn velocities, and upper‐mantle structure that affects travel times at regional distance, mostly from a scaled 3D S‐velocity model for the same region. This fact may imply that although errors may be included in this scaling, this way would work better than conventional P‐arrival inversion. This difference likely exists because surface waves have a better lateral resolution for the crust and uppermost mantle than travel times.
Online Material: Histograms of travel‐time residuals and correction surfaces of travel times with respect to iasp91 for 33 stations.