Algorithms for locating earthquakes recorded by local networks commonly assume a flat-layered velocity structure, even in island arc regions where the presence of a dipping lithospheric slab introduces large lateral velocity variations. Moreover, because seismic stations occur along the narrow island arc, the resulting linear network geometry degrades earthquake location. To model this situation, we trace rays in a spherical earth from hypothetical earthquakes through appropriate velocity models to networks of various geometries. To isolate the effect of the slab, most of the velocity models consist simply of a planar-dipping constant velocity region embedded in a sphere with a constant lower seismic velocity. We then use the calculated P and S travel times to relocate the hypothetical earthquakes with a flat-layered velocity structure. We perform calculations of this kind in a variety of models, varying the dip of the slab, the velocity contrast between the slab and the mantle, the station geometry, and the distance between the earthquakes and the slab-mantle boundary. We also perform calculations in more realistic models with geometry closely resembling two seismic networks in Alaska.
These model calculations emphasize the great care that seismologists should exercise when interpreting local network locations in island arc regions. In the present study, all the relocations incorporate P and S arrivals at all stations, and include no random or “picking” errors. Nevertheless, the relocated hypocenters differ from the “actual” model hypocenters by up to 100 km or more and produce significant distortions in the Wadati-Benioff zone. The most prominent spurious feature is a clear increase in the apparent dip of the Wadati-Benioff zone beneath a certain depth. Furthermore, the relocation reduces the width or apparent thickness of the Wadati-Benioff zone, e.g., they produce an apparent thickness of 5 km or smaller for a zone of earthquakes with an “actual” thickness of 20 km. However, if different combinations of stations report different events, it is possible to determine an apparent thickness which exceeds the “actual” thickness.
For local networks near subduction zones, it is useful to estimate the mislocation and rms residual expected for each event recorded and located. This can be accomplished by performing model calculations in a simple model with a constant velocity, dipping slab embedded in a half-space. For such a simple model, there exists only one ray path from each earthquake within the slab to each station, and the travel time for this path can be calculated as a part of the routine network location process. A relocation using the routine network location algorithm and these calculated travel times provides an indication of the expected mislocation and rms residual for the actual recorded event.